Amman-Iar stress & gemination

Before beginning, I'd like to observe how living with the same conlang for
many years gives one native-speaker intuitions. I'm reminded of Tony Harris
(a much-missed former presence on the list) who had very little linguistics
knowledge, yet wrote his personal journal in Aluric.

There's tons & tons of quoting in the following message, because it's
such a good thread. We don't often get the chance to analyse one another's
conlangs.

OK. The facts of Amman-Iar stress:

David:
>in amman iar the stress is always penultimate in 2-syllable words, as in
>niran [nir'an].  In longer words, it occurs penultimately if that syllable
>contains a vowel followed by two or more consonants as in orathval
>[orrath'val] or a diphthong as in erainin [erain'in].  If the penultimate
>syllable contains a simple vowel followed by a single consonant or another
>vowel, the stress falls on the syllable before it as in tarnarin [tar'narin]
>and nimroin [nim'roin].  Note that the digrams ch, th, dh, and sh represnt a
>single consonant, tus erathin [er'athin] not [erath'in].  Stress in amman
>iar is basically quantitative, i.e. syllables are distinguished by vowel
>length.
Matt:
> Aha!  The old Latin/Quenya stress rule!  A simpler way to state it would
> be:  "Stress the penultimate syllable if heavy, otherwise stress the
> antepenult."  Amman Iar, like Tokana and numerous natlangs, appears to
> distinguish 'heavy' syllables (those ending in a consonant or diphthong/
> long vowel) and 'light' syllables (those ending in a short vowel).
David doesn't specifically say how C + liquid clusters are handled. In
Latin, of course, they form a branching onset, so the liquid does not
affect the stress. In Quenya, the picture is less clear; it was debated
on Tolklang some years ago and IIRC (which I very very probably don't) it
seemed to be the case that JRRT stated the rule like David, but applied it
like Latin.

David:
>Case endings (with notable exceptions among pronoun) cause gemination of the
>final root consonant and thereby shift the stress, e.g. adhan [adh'an],
>adhan+ERG > adhan+e > adhanne [adhann'e].
Matt:
> I wonder if maybe the stress shift is the primary change here, while
> gemination is the concomitant change:  Case endings trigger a shift in
> stress to the final syllable of the root.  Amman Iar metrical rules
> prohibit light penultimate syllables from carrying stress, and so the
> final consonant of the root undergoes gemination in order to transform
> the penultimate light syllable into a heavy syllable.
>    If I were doing a linguistic analysis of Amman Iar, I might be tempted
> to analyse gemination this way.  I guess the question to ask is:  What
> happens if the root ends in a diphthong or consonant cluster?  Or do
> such roots not exist?
David replies:
> I love it!  I absolutely love it!  Leave it to Matt to come up with a better
> analysis of my conlang than I have myself.  In fact, while roots ending in
> diphthongs don't exist, those ending in consonant clusters do and lo and
> behold, they don't geminate, thus sarn+DAT > sarn+en > sarnen not *sarnnen.
> Under my analysis these had to be dealt with as exceptions, under yours,
> they are quite regular.  Occam's Razor at its finest.
He then adds:
> On reflection, this is not quite correct.  The heavy/light syllable
> distinction is not really operative here, although my previous description
> did not really adequately reveal that.  Thus
>    1 [C]VC VC' CVC erinvar [erin'var]
>    2 [C]VC VCC' VC orcarmar [orcarm'ar]
>    3 [C]VC' VC VC erinis [er'inis]
> In each of the above, the penult is 'heavy', but in (3), the antepenult
> takes the stress.  (1) and (3) show that it is not the quality of the penult
> (whether light or heavy) in isolation that determines the stress pattern.
> Apparently, the mere existence of a coda in the penult is insufficient to
> trigger stress.
I would say that it is the stress that diagnoses syllabification. You
may intuit the chunking you describe, but you are perhaps describing
morphological units rather than phonological ones.

Hence _er'inis_ would phonologically chunk as _e-ri-nis. Matt was asking
about your native speaker intuitions about this, and taking them to be
decisive, but while your intuitions are valid, your intuitions about
what they are intuitions of (i.e. phonology versu morphology) are not
(I declare, speaking as a linguist).

Matt elaborates:
> Dirk wrote:
> >The preference for syllables to coincide with morpheme
> >boundaries seems also to be operative in Amman Iar, but in that
> >language, the resolution of the conflict is different. Rather
> >than delete consonants which find themselves on the wrong side
> >of the morpheme=syllable boundary, Amman Iar instead prefers to
> >violate NoCoda; that is, the congruence of syllable and morpheme
> >boundaries takes precedence over an unmarked syllable structure.
> >
> >I think this is rather cool, myself.
>
> So do I!  So Dirk:  How can we reconcile all of this with the
> tantalising similarities between the Amman Iar stress rule
> and the Latin stress rule, as well as the operation of the
> Amman Iar gemination rule?
>
> How about this: Inflected words in Amman Iar are initially
> syllabified in accordance with NoCoda and other constraints
> which enforce an unmarked syllable structure.  Stress
> assignment and gemination then operate on the basis of
> that representation.  Finally, a "syllable-boundary
> readjustment" rule (SBRR) is applied, which reassigns
> certain onset consonants to coda position, in accordance
> with a constraint which enforces congruence between
> morpheme and syllable boundaries.  A sample derivation
> would look like this:
>
> Take a noun like "adhan" = "man":
>
>        'a.dhan
>
[1]
> Adding the ergative suffix "-e" triggers a stress shift
> to the right, together with resyllabification:
>
>        a.'dha.ne
>
[2]
> Because of a constraint against stressed penultimate
> syllables being light, gemination takes place:
>
>        a.'dhan.ne
>
[3]
> Finally, the SBRR is applied, yielding the correct surface
> form:
>        a.'dhann.e
>
> Something like this might work, yes?
I don't believe [3] is necessary, because there is no linguistic
evidence for it.

My objection to Matt's analysis [1-2] is that it must be *stipulated*
that it is gemination that resolves the conflict between the
stress on the penult in "a.'dha.ne" and its lightness. This is
because there are other ways the conflict could be resolved, such
as by inserting some other consonant, by lengthening or
diphthongizing the vowel, or by reversing the stress shift.
You thus need two stipulations: (a) "-e" triggers stress shift; (b)
resolve conflict by gemination.

A version of David's original formulation is in fact simpler and more
explanatory. That is, there is a single stipulation: (a) "-e" triggers
gemination. The stress shift follows from (a) without stipulation.

I suggest that the actual rule is:

   Case endings must follow a branching rime. [= a heavy syllable]

The suffix "-e" is therefore specified thus:

    R   O + R
   / \  |   |
  X  X  X   X
            |
            e

where R = rhime, O = onset, + = morpheme boundary. I assume that
codas are possible only before onsets.

"Adhan" is thus:

   R   O   R   O
   |   |   |   |
   X   X   X   X
   |   |   |   |
   a   dh  a   n

When combined with the pattern specified for "-e", you get:

   R   O     R  O + R
   |   |    / \ |   |
   X   X   X  X X   X
   |   |   |    |   |
   a   dh  a    n   e

By a general rule of filling empty structural positions by spreading,
you would get:

   R   O     R  O + R
   |   |    / \ |   |
   X   X   X  X X   X
   |   |   |   \|   |
   a   dh  a    n   e      = a.dhan.ne

Stress by the general rule would fall on the penult.

For bases where the final rime was already branching (i.e. ending in
diphthong or cons cluster), there would be no empty position following
attachment of the affix, and hence no need for geminate-creating spreading.

--And.