Re: OT: What makes a good conlang? (was Re: Super OT: Re: CHAT : JRRT)

Trebor wrote:

<<As an aside, which is more reasonable (or are they all possible)?
t -> l
t -> h
t -> ?>>

Simple answer: /t/ > [?].   However, for this to be a unilateral change would
be very strange.   This is a sound change of English, though.   Final /t/
became an unreleased /t/ (don't know the X-SAMPA) which then became a glottal
stop (in some dialects, and not all the time), in words like "cat", "bat", "rat",
"hit", etc.   However, it also happened in my dialect before a syllabic
nasal.   So, "kitten" is now ['k_hI.?n=] in my dialect.   Lots of people have
['k_hI.4@n], though (which sounds like toddler talk to me, but it's no more
unreasonable then the sound change that occurred in my dialect--actually, maybe even
more believable).   Oh, the environment isn't the same, though: Where flapping
occurred, the nasal is not syllabic.   That could be the difference.

Anyway, /t/ going directly to [h] is totally unrealistic.   I can't see it
happen.   However, /t/ can go to [T] (hopefully not in all environments), which
can go to [f] which can go to [h] (or the [f] step could be skipped).   So
that's one way to get a /t/ to an [h].   However, you have to consider that if
this sound change is happening to /t/, similar sound changes are probably also
happening to your other voiceless stops.   This could result in /p/, /t/ and
/k/ *all* going to [h], and you certainly don't want that.

Now, /t/ to [l] is unrealistic, but I can't as yet say why...   First of all,
your voice feature is changing, and your lateral feature.   Also...hmm...
This directionality just doesn't work.   I've seen plenty of /l/ -> [d] sound
changes, and /l/ changing to lots of other approximant like things, but never
/t/ -> [l].   It could work, though, if you had a flapping rule, like in
English, that made /t/ -> [4], let's say intervocalically, and then you could have a
merger where /4/ merged with [l].   In that way, a /t/ could kind of become
an [l].

The best way to learn about reasonable sound changes is to see what kinds
have occurred in natural languages throughout the world.   This will give you an
idea of what is statistically likely.   It will not, however, teach you
anything about motivation.   Motivation comes from acoustics, generally, but can
also be inventory driven.   For example, take the famous Hawaiian pull chain.

Hawaiian used to have the following (just looking at stops and fricatives):

/p/, /t/, /k/, /?/, /h/, /s/

Then something happened, and all glottal consonants were lost, these being
/?/ and /h/.   You can imagine this happening fairly easy, as it's happened in
lots of languages (Spanish, French and some dialects of British English, being
some that come to mind for [h] loss).   This left Hawaiian with the following
inventory:

/p/, /t/, /k/, /s/

The void left by the glottalic consonants, however, created a kind of black
hole, and it started to suck nearby things into it.   So, for example, the only
fricative that there was was changed into an /h/:

/s/ -> /h/

Then, for the glottal stop, /k/ was taken:

/k/ -> /?/

That left a void at the velum, which wasn't acceptable.   So, the /t/ was
dragged:

/t/ -> /k/

This leaves a void in the alveolar region.   *However*, Hawaiian still had
/l/ and /n/.   The only labial element (aside from /w/) was /p/.   The only
thing that could've been dragged to fill the /t/ void was /p/, but that would have
been too drastic, so /p/ was left in place, leaving Hawaiian with the
following inventory for stops and fricatives:

/p/, /k/, /?/, /h/

This is one way that a sound change can be motivated.   Deciding that random
segments should change to other random segments is generally unmotivated
(though it obviously does happen).   But let's take your /t/ -> [h] example.   It's
really strange, but take Finnish, where a coda /k/ before another stop
becomes none other than [h].   If this happened not just to velar consonants, but to
all stops, then there's a situation where /t/ could become [h].   This would
apply only in a restricted environment, though, so you'd still get /t/'s
elsewhere.

A kind of shortcut to learning what kind of changes are more natural is to
learn featural phonology.   This is where every sound of a given language is
specified as either [+] or [-] some feature.   So, take /t/ for example.   /t/
is...

[+consonantal]
[-voice]
[-vocalic]
[-sonorant]
[-nasal]
[-continuant]
[+coronal]
[+anterior]
[-retroflex]
[-lateral]

I just made up that last one (also, I'm not sure if "anterior" is used
anymore).   Anyway, a natural sound change is one where the fewest number of
features are changed, and this usually happens becomes something in the environment
is affecting it.   So, let's say you wanted to have /t/ voice to [d]
intervocalically.   Then, all you'd be changing is one feature...

[+consonantal]
[+voice] <------------- Changed
[-vocalic]
[-sonorant]
[-nasal]
[-continuant]
[+coronal]
[+anterior]
[-retroflex]
[-lateral]

Also, notice that if it's in between two vowels, the two vowels are specified
as [+voice].  So you could say that the voice feature is spreading from the
vowel to the consonant, and that's why it voices.   This gets you into the
realm of feature geometry, which was all the rage before OT thrust it's big, ugly,
ill-conceived head into the scene.

Now, let's consider your change of /t/ to [l].   We know the features of /t/,
so let's see the features of [l], and how many changes it involves:

[+consonantal]
[+voice] <---------------- Change 1
[-vocalic]
[+sonorant] <----------- Change 2
[-nasal]
[-/+continuant] <------ Change 3*
[+coronal]
[+anterior]
[-retroflex]
[+lateral] <--------------- Change 4

*Some linguists contend that laterals are [-continuant].

This is four (or three) major changes, none of which are motivated.   This is
a kind of phonetic explanation for why /t/ to [l], without any context, is
unrealistic.   Then /t/ to [h] is a huge change, since you're changing the place
node, as well as the [spread glottis] feature (provided the /t/ is
unaspirated).

Now let's take a look at another sound change you proposed /h/ -> [j].   Here
are /h/'s features:

[+consonantal] (debatable)
[-voice]
[-vocalic]
[-approximant]
[-sonorant] (apparently debatable)
[-nasal]
[+continuant]
[-labial]
[-coronal]
[-dorsal]
[+spread glottis]
[-constricted glottis]
[-high]

Some have debated whether or not /h/ and /?/ should be considered
consonantal.   I think the verdict was it depends on the language.   I assume if you want
to treat this thing like a--well, actually if you want it to change to [j],
it might be better to specify it as [-consonantal], so let's do that.   Also,
it's apparently debatable as to whether /h/ and /?/ are [+sonorant] or not.   I
would've said [-sonorant], without a doubt, but my chart says [+sonorant].
I don't know if I trust it.   Anyway, now let's count the changes from /h/ to
[j]:

[-consonantal]
[+voice] <---------------- Change 1
[-vocalic]
[+approximant] <------ Change 2
[+sonorant] <----------- Change 3 (unless you count an /h/ as [+sonorant])
[-nasal]
[+continuant]
[-labial]
[+coronal] <------------- Change 4
[-dorsal]
[-spread glottis] <------ Change 5
[-constricted glottis]
[+high] <----------------- Change 6

So there are a bunch of changes here (and possibly more, considering that
I've never seen a convincing argument for /h/ being considered [+] or
[-consonantal]--same goes for [j], too).   Even in the best possible situation, you'd
still have five changes (*maybe* four, if you get away with calling [h] an
approximant), which is way more than you'd want in one jump.

I kind of think of /h/ and /?/ as the end of the line.   Once a phoneme has
evolved to either /h/ or /?/, all it can do is disappear.   Maybe Teoh might
consider a pull chain of some kind...?   That'll leave you with some blank spots
somewhere, but that might be interesting in and of itself.

-David

Reply