Dick:
> And to me:
> >### I'm afraid I have to confess here that I did stray into phonology in my
> >#WG lectures last term. I suggested that phonology has a dependency
> >#structure just like syntax
> >
> >This much should not be too controversial, in that it boils down to the
> >claim that syntagmatic structures are headed, which would, I think,
> >be widely tho not universally accepted.
> >
> >#and sounds are classified just like words, by
> >#multiple default inheritance.
> >
> >This is more controversial, since it seems to imply the view that
> >segments are minimal elements of phonological structure, rather
> >than structured composites of subsegmental elements.
> ## What does it mean to say that a subsegmental element is different from a
> class of sounds? E.g. if I say that /b/ isa {voiced, plosive, labial}, each
> of which is a sound type, would that be different from saying that it
> consists of three subsegmental elements (given that these elements occur
> simultaneously, not one after the other)?

[please take as read some disclaimer from me about not being a phonologist]

Here are some problems with the classificatory approach:

1. In Government Phonology and Dependency Phonology one subsegmental element
is the head of another, and this head--dependent relation is different
from those that build syntagmatic structure only in that the subsegmental
elements are realized simultaneously.

2. In Particle Phonology, multiple instances of a subsegmental element type
may be present within a single segment.

3. Even in more traditional feature-based models, features are treated as
autonomous elements that can move and spread. Only with considerable
contortion could this be modelled in terms of features being category
memberships, and even then the analogy with syntax breaks down, because
although feature spreading (= harmony, assimilation) resembles a
feature-sharing account of grammatical agreement (which I am antagonistic
towards, I might mention), there isn't a grammatical analogue of feature
movement (here I ignore howls of disagreement from P&Pers).

4. Where does the classificatory approach stop? Why not classify syllables
rather than subsyllabic segments? If your miniminal units must have the
property of being single phones, then why not classify long monophthongs
that pattern with diphthongs and VC rhymes as Long (which is unexplanatory)
rather than as a sequence of two vowels?

> >#It was mainly in the context of morphology,
> >#so I didn't really get into intonation (though Yoshi tried to push me in
> >#that direction!). One of the attractions of a dependency phonology is that
> >#it explains why past tenses are made by changing the stem vowel (ring -
> >#rang) - this is the bit of the phonology which is directly connected to the
> >#stem
> >
> >I'm not sure we want an explanation for this. IIRC, its origins go back
> >to Proto Indo European ablaut, so the synchronic question is whether
> >the language can have stem alternations of any sort, or whether there
> >are constraints on possible stem alternations. Since stem alternations
> >in English form a closed and dwindling class (except for the occasional
> >analogical extension of the synchronic ablaut pattern), it seems to me
> >that the most economical analysis is one that simply lists the alternations,
> >without invoking additional machinery that will then 'explain' the
> >alternations.
> ## Fair point; moreover vowel alternation in semitic-type languages affect
> all the vowels, not just the stressed one. (I should know - that's how Beja
> works.)
>
> >
> >#(in much the same way that the sentence root is the only word that's
> >#directly linked to meaning, since it carries the whole sentence meaning and
> >#all the other words just help it along). If only someone more competent
> >#than me would do some serious work on phonology in WG ...
> >
> >I don't see why this is so desirable. As I said in a recent message, a
> >theory is made up of foundational assumptions, of a representational model,
> >and of specific analyses. Specific analyses are usually not intrinsically
> >associated with a theory, cand can be translated from one to another.
> >And I don't think there's any reason to doubt that WG's foundational
> >assumptions are incompatible with phonology,
> ## What a splendid sentence!! If I've counted the number of negatives
> correctly, it's odd, so we should give up trying to make WG work for
> phonology. Right?

Whoops! Read: "And I don't think there's any reason to think that WG's
foundational assumptions are incompatible with phonology" or "And I don't
think there's any reason to doubt that WG's foundational assumptions are
compatible with phonology".

-- Another reminder about the difference between logic and the mental
resources that we use to implement it...

> so we don't need 'WG
> >phonology' for this. So all that remains is the issue of whether WG's
> >representational model can extend to phonology. I'm sure that any
> >current model of phonological representation could be translated
> >into a WG conceptual network representation (-- the same goes
> >for syntactic models),
> ## I'm not so sure. Some people think phonology is procedural, and I don't
> think procedures can be represented in a network (except declaratively, of
> course).

Good point. I don't know whether anybody thinks that procedural models
are actually modelling processing per se, but certainly the vast majority
of theories formulate their models procedurally and indeed the motivation
to do so is rather more compelling than it is with syntax. But that said,
I believe any procedural formulation can be translated into a declarative
one. (I think most syntacticians except Chomsky agree, but as I understand
it, this is still a contested issue in phonology (somewhat to my
bemusement).)

> I doubt if OT translates into WG.

I see why you say this, but it would be strange if it were really true,
what with OT being just a theory of ranked constraints rather than of
procedures. I strongly suspect that you could use inheritance
to model OT's ranked constraints -- in both cases the key instrument
is overriding.

> so this leaves,as the one open question, the
> >issue of whether the specific mechanisms of syntactic structure carry
> >over to phonological structure, and I don't see why this issue should
> >matter so much to WG.
> ## Well, it matters to me because I think all knowledge is unified, so
> either phonology fits into the same network, or we have a serious interface
> problem.

This is a different question. It could be possible to model phonology
using WG's conceptual network formalism, and to relate phonological
structures to words by means of relations such as Whole-of, but for
the specific mechanisms of syntactic structure to have little in
common with the mechanisms of syntax. (After all, if syntax is the
servant of meaning, and phonology is the servant of phonetics, in
principle syntax and phonology could be as different as meaning and
phonetics are.)

--And.