Re: Math/Phonological formulae

Roger replied to a message of mine, but he was over the 5 message
limit.  O Master of Instrumentality: Does this mean that his message
will reappear automatically after 24 hours?  Roger seems to think so;
I can't remember what the rule is.  Based on Roger's assumption,
though, I'm simply going to reply to it, and not quote the whole
thing, on the principle that we will be able to see it some time in the
near future.

Roger:
<<
> No way cheating!!! :-) It's simply an abbreviation of the
> theoretically
> full form of the rule:
>
> [+cons (all necessary features) +lat +voi] --> [+cons (all
> nec.features)
> +lat -voi] (no env.bar is necessary).
 >>

No, you were referring to the wrong part of the rule.  Here's the
rule you wrote again:

[+voice] --> [-voice] / [+lateral ___]

I was saying that the [+lateral] was cheating.  Actually, I don't
think the rule as it's written makes sense...  What corresponds to
the _?  A feature?  If so, which feature?  What I was saying is
that shouldn't the [+lateral] be in the first feature matrix (the one
with [+voice])?  And then shouldn't the environment be different?

Roger:
<<
> If all, or certain groups, of C devoice in a given env., that could be
> handled (I think), but it might be a very complicated statement
> (probably
> with lots of Greek-letter variables, which I haven't gone into).
 >>

Duh!  Geez, not even a year out of graduate school, and I've already
forgotten the basics!  Yes, Greek variables make everything easier
(though they do make it look like there's a connection where sometimes
there isn't).  Hard to type in ASCII, though.

Roger:
<<
> Unless you're working with some other theory, features are _by
> definition_
> binary in Jakobsonian/C&H/orthodox Gen.Phon. Multiple-value
> features have
> been proposed (in fact, they work well for stress), but aren't
> generally
> accepted. Nor can you have a binary feature without a +/- sign.
 >>

Referring to the features CORONAL, DORSAL and LABIAL. Yes,
this is a different theory--by most accounts, an updated version
of the theory.  I believe in the theory you're working with, you
have...

[+front, -high] = labial
[+front, +high] = coronal
[-front, -high] = palatal (?)
[-front, +high] = velar/uvular

Something like that?

In the updated version (and I don't know who was responsible),
it was reasoned that the place features (this almost looks like the
place nodes of feature geometry...) don't have +'s or -'s: they're
just places.  So you'd say it's CORONAL, and then within the
coronal realm, you'd say whether it was +/- anterior, to distinguish
alveolars from alveo-palatals and palatals.  DORSAL, then, has
it's own features, as does LABIAL, and the features only apply
within those domains.  So something can't be DORSAL and +ant,
because [anterior] isn't a feature of dorsal consonants.  This does
kind of break down the whole binary features thing, which, if
it's considered as vital as it has been, calls into question whether
someone who uses DORSAL is even operating within the theory.

Roger:
<<
> I think your rule is a case where Greek-letter variables might
> work-- as I
> understand your rule "a C that is +ant and EITHER +cor (t,d) OR +dor
> (k,g)" -->> [+cor -ant] etc.
> OK, let's see (using a for "alpha"):
>
> [aCor -aDor +ant] --> [+Cor -Dor -ant ] / [+Cons ___] [+V [-back,
> +ATR]
 >>

I must admit, you're using some kind of notation that I'm not
familiar with, because I've never seen the ___ used within a
feature matrix.  But this is the problem one runs into with the
non-binary feature version of the theory.  COR and DORS can't
have +'s or -'s, so they can't have alphas...  I guess since in my
inventory there are no [+ant] non-coronals, I can take out COR,
but that leaves [+ant] and [DORSAL]...  If I had the [back] feature,
I could do [aant, -aback] > [-ant, -back].  Hmm...

Roger:
<<
> But if both td and kg > c (whether [tS] or [ts], which I assume is
> meant)
 >>

You assume too much, I'm afraid.  What I meant by [c] was...[c]:
a voiceless palatal stop.  The palatals are:

[c], [J\], [C], [j\], [J], [j] and [L]

Roger:
<<
> then your +cor td have not actually changed w.r.t. coronality. (You do
> need to add a feature for the afficate quality, +strident is
> common, or
> you can invent one, as I have for Gwr-- [delayed release]. Or am I
> missing
> something????)
 >>

The COR feature does not, indeed, have to change for coronals--
I realized that above.  Since they're true palatals, though, I don't
have to worry about affricates, so...yay!

Roger:
<<
> I think my "high" corresponds roughly to your "dorsal" though not
> entirely; I also use "front" for your "anterior", so that /pb/ are
> +fr,-hi; /td/ are +fr, +hi, /kg/ are (-fr)+hi,+back. Jakobsonian
> features
> differ from mine, also from C&H; Jak. +grave (and other features)
> covers
> pb/kg, which is helpful in some cases...)
 >>

Oh, so this explains.  Wow, so it has [front] *and* [back] as features?
Now, that *is* cheating.  ;)

Roger:
<<
> Well now you're in the "underlying form" dilemma-- how
> "abstract/historical" are underlying forms? The proto-lang. had no
> palatals; the modern lang. does. If palatalization is absolutely no
> longer
> active in the modern lang., then _modern_ underlying forms will have
> palatals (in effect, the morpheme structure rules have changed in the
> transition from proto to modern-- that's very common).  OTOH, if
> palatalization is still an active rule, then your underlying forms may
> have to be more historical in appearance.
 >>

The palatalization is still active, in a sense.  So if you have a
noun like
[cet], there's no way to know whether it was originally *ket or *tet--
that's opaque.  However, if you add a suffix (like /-en/), you get
[cecen].  This variation a speaker can pick up on easily.  Oh, I should
read ahead before responding...

Roger:
<<
> Or again (on the third hand), it
> may be that palat. has operated fully on proto-forms, but now
> sequences
> can arise in the modern lang. where it may apply again--  in that
> case,
> palat. might be a relatively late rule in the underlying>>surface
> rules.
 >>

Yes, this is the case (as illustrated above).

So here's the question: Given that state of the language, let's say
that,
for whatever reason, a speaker wants to coin a brand new word with
a brand new phonological form (it happens.  Think about product names.
Clorox had to come from somewhere).  Will that speaker be able to
produce forms like:

[cat]
[Col]
[Jun]

?

That's the question I've been struggling.  Because, if this is the case,
then perhaps new words should have sequences of palatal consonants
followed by non-palatalizing vowels.  If it's not the case, then they
shouldn't.  That's the dilemma.

Roger:
<<
> [wVtS@...] [dZitSEt] ("what are you...?, did you eat yet?" etc.)
 >>

That second example is priceless, so I'm repeating it for posterity.  :)

Roger:
<<
> It can be done, simply add -str to the left-hand side, though the env.
> might then need more specificity.
 >>

This was regarding my stress rule.  I can write it easily enough, but
I can't *type* it.  It's two hard to write syllables and indicate stress
in an unambiguous way.  Lousy ASCII!  When are we going to be
able to just write on the screen with a pen and have it accept that?

Also, responding to Dirk...
<<
Ranting aside, whenever *any* theory takes precedence over the
description and explication of natural language phenomena, it has
sidetracked linguistics. OT provides some insight into the interaction
of real phonological patterns, and as such, it is a useful tool.  But
it isn't (or shouldn't) be the only tool in the toolkit. It's only now
that "mainstream" phonologists are coming to grips with things like
gradience, variation and frequency effects--which wouldn't have
happened if it weren't for OT.
 >>

I certainly recognize that it has done some useful stuff (I like the
way it's been used to conceptualize stress placement).

But the problem I ran into (and the reason that I left phonology
for good) is that, theory aside, if you sit down to describe the
phonology of a language, what do you do?  I've come up with
a way that I do it in my conlangs that's satisfactory to me, but
within linguistics?  Do you combine templates with SPE rules?
Feature geometry?  If not, do you decide on one theory, come
hell or high water, and try to force everything into it?  It seems
to me that the answer is the latter, and that that's probably not
a good way to go about it.  Nevertheless, you have to be able
to say something in some way, and if it's not within a single
framework...you're kind of stuck.  That's the way I always felt,
anyway.

-David
*******************************************************************
"A male love inevivi i'ala'i oku i ue pokulu'ume o heki a."
"No eternal reward will forgive us now for wasting the dawn."

-Jim Morrison

http://dedalvs.free.fr/

Replies