Hallo conlangers!

On Thursday 16 August 2012 00:29:17 BPJ wrote:

> On 2012-08-15 19:22, Jörg Rhiemeier wrote:
> >> >My perception, however, is that conlangers have a tendency
> >> >to have a "grand master plan" and apply the sound changes
> >> >more or regularly; but the more experience (e.g. Tolkien)
> >> >will build in naturalistic irregularities.
> > 
> > Yes.  I have to admit that I used a GMP in Roman Germanech, but
> > even there I sometimes "bent" a word that fell out of the GMP
> > and did not feel "right".  For Old Albic, I have an outline of
> > the sound changes from Proto-Hesperic to Old Albic, but I do
> > not treat these sound changes mechanically!
> 
> Yes, you have to have a plan -- at least in your head --,
> but only as a general guide for the regular case. IMHO only
> the human mind and eye can *effectively* spot where
> irregular changes like metathesis or analogy would be likely
> to occur, by examining all the intermediate stages.

Exactly.  The human mind is flexible enough to handle such
irregularities; computers are not.

> I usually follow the plan rather strictly at first, but if
> I spot a possible irregularity or analogy along the way I
> make at least a mental note of it and if the regular outcome
> somehow rubs me the wrong way I can go back and follow
> the path with the irregularity.

With Old Albic, I have often found that applying the sound
changes I have to a root yields a result that rubbed me as
"wrong", and I decided how the word "should" sound - and
added another sound change rule to the list to account for
it.
 
> > With none of these languages, I feed lists of words
> > and sound changes into a sound change applier
> > program.
> 
> I wouldn't say that it's categorically wrong to do so, and
> I have done so in the past.  However you need
> a sophisticated program and sophisticated rules to get
> sophisticated results, and to write a sophisticated program
> and/or rules which accurately describe your soundchanges in
> a way which the program can digest and deploy effectively
> and accurately you need a lot of time and energy.

Indeed, this is the main reason why I don't use those programs
- they are simply not flexible enough for my purpose.

>       The human
> brain comes with most of the needed sophistication,
> including fuzzy logic and dowhatimean capabilities, which
> one can make a computer match only with much hard work.
> Imagine supplying a program with the amount of panoptic view
> of the vocabulary and fuzzy matching capability needed to
> spot potential analogies!  Or just to calculate the
> interaction of rules.  The human brain OTOH is good at those
> things 'out of the box'.  You simply get more conlanging
> done in the time you spend actually doing the work in your
> head as opposed to spending time setting up the
> automatization!

Very much so - and actual conlanging work is, IMHO at least,
much more fun and rewarding than getting a sound change
application program to work properly!

>       Also any revision of the rules gets
> complicated with a SCAP.  I found myself refraining from
> introducing revisions I vanted or needed, because I shunned
> away from the rewriting of the input to the program, and in
> the end I abandoned the program!

I can imagine that very well.
 
> Even when I did use a program I never trusted it blindly:
> I made sure it logged all intermediate stages down to minute
> detail, and always went through the log and made hand
> tweaking as I saw fit.  I'd often get sleepy during those
> review sessions, but I never do when doing the actual work
> in my head/on paper -- just like proofreading is tedious and
> 'sleepy' work, as opposed to writing and normal reading!

Sure!  There is little point in automatizing a process such as
the development of the vocabulary of an a posteriori conlang
if the automatization actually requires as much work to assure
that the automaton ran the right course, as one would have if
one did the work "manually"; and as you say, actually making
up things is more fun than checking whether a machine has done
the right things or not!
 
> It's the same with random word generators.  For me they are
> a necessary evil in apriori conlanging, since what I manage
> to pull out of thin air is usually repetitive both within
> and across the shapes I come up with.

I find a priori lexicon development very difficult, too.  One
of the reasons why Old Albic is partly a posteriori.  (The main
reason for a posteriori items in the vocabulary of Old Albic is
of course that I imagine it to be related to Indo-European and
to the source languages of potential substratum loanwords in
western European IE languages.)  Yet, I haven't used a random
vocabulary generator at all so far, because I want the words
to sound "right", and I feel that a machine cannot capture that.

>       However I never
> randomly generate a huge random vocabulary for a huge list
> of glosses for serious projects.  I always have the
> generator suggest a number of words -- usually five to seven
> -- at a time, and if none of them seems to fit the meaning
> I have in mind I call up a new set of suggestions More often
> than not the generated forms help me to know what the
> fitting form should *not* be like, and I come up with
> something fitting spontaneously, and in any case I usually
> only let the generated forms guide me: the form I actually
> pick is mostly not exactly like any one generated form.  So
> it's 'computer suggested' conlanging rather than automated
> conlanging.  But still I've spent an inordinate amount of
> time writing RWGs!  For me that's less of a waste of time
> though, since my brain ain't good at spontaneous logogenesis
> -- anymore; you should have seen me when I was twelve! :-)

I also find coining new words spontaneously difficult, but so
far I never felt that a random word generator would make my job
easier without degrading the quality of my work.  After all,
that beast needs to be programmed and tested, which takes up a
lot of time and effort which I could spend in a more entertaining
way on the language itself!

On Thursday 16 August 2012 05:42:15 Alex Fink wrote:

> On Wed, 15 Aug 2012 19:22:37 +0200, Jörg Rhiemeier <[log in to unmask]> 
wrote:
> >On Wednesday 15 August 2012 17:05:30 R A Brown wrote:
> >> Well, no - things like metathesis happen  ;)
> >> 
> >> Sound changes in natlangs do not always happen regularly.
> >> Metathesis is one feature that upsets this, e.g. from Latin
> >> _perīclu(m)_ one would expect Spanish for "danger" to be
> >> *periglo; it's not - it's _peligro_.
> >
> >Right.  Such things often happen irregularly.
> 
> But, to be clear, metathesis _can_ be regular!  For instance, in parts of
> Slavic, a coda liquid regularly metathesised with the preceding vowel; in
> Chukchi, a coda /q/ regularly metathesises across the vowel to become an
> onset /?/.

Yes, metathesis can operate regularly; and I have some regular
metathesis in the diachronic development of Old Albic (for
instance, *TVRC regularly becomes TRVC where T is a stop and R
a liquid).

> See e.g.
>   www.ling.ohio-state.edu/~ehume/papers/hume_metathesisS5.pdf
> and what it says about the "metathesis myth".
> 
> >I've heard about programs which apply automatic sound changes, but I don't
> >understand how this is supposed to work with all the other stuff going
> >on, like the analogical levelling you mention, or the fact that words get
> >replaced, or are influenced by other words to behave irregularly.
> 
> Yes, I also haven't actually used a sound change applier to do my
> conlanging except as a proof of concept device: I'll use them say to apply
> the first batch of sound changes to passages in the source language and
> look at the result to decide what subsequent sound changes would be
> natural, because I'd created some marginal contrasts or phonological
> imbalances or ... without realising it.  But then I do do the real work by
> hand for the analogy and the one-off influences and stuff.

One should not apply too many sound changes in one step, no
matter whether one is doing that manually or automatically.
You often need to develop a feeling for an intermediate
stage before you can decide in which direction to move on.

If you don't check intermediate stages, you are running the
risk of buggering the thing up - e.g., by eliminating too
many distinctions and coming up with a phonology from which
the desired endpoint is hard to reach.
 
> I'm a little suspicious of doing the last stage by the gut -- even though
> that's what I do -- 'cause it seems like a way for things like one's
> native-language phonaesthetics and the like to sneak in.  But I guess you
> could be careful and explicit at some level about this too.

Yes.
 
> Anyway, in the large, I think conlangers as a whole are pretty good at
> sound change and at the broad-brush type of morphological change
> (grammaticalise tense! get rid of case!).  I've seen many fewer well-made
> examples of e.g. gradual slight reworking of morphological paradigms, or
> the sorts of morphological evolution which lead to conglomerately forming
> new affixes or morphological excrescence or stem extension settling or
> these kinds of phenomena (not from myself either).  Morphological analogy
> can act in diverse and subtle ways.

Some diachronic conlangers tend to neglect morphological change
as against sound change.  They just apply the sound changes (by
hand or by machine) and accept what they do to the paradigms.
Real language change doesn't work that way.  Rather, sound
changes trigger changes in the morphology, be it analogical
restoration of lost regularities, abandonment of categories due
to phonological loss or "bleaching" of the relevant morphemes,
or grammaticalizations of derivational forms to make good for
such losses.

--
... brought to you by the Weeping Elf
http://www.joerg-rhiemeier.de/Conlang/index.html
"Bêsel asa Éam, a Éam atha cvanthal a cvanth atha Éamal." - SiM 1:1