On Wed, 27 May 2009 13:52:45 -0700, Sai Emrys wrote:

> On Wed, May 27, 2009 at 1:25 PM, Jörg Rhiemeier <[log in to unmask]>  
> > There are many things that occur in only very few natlangs and
> > are nevertheless perfectly naturalistic, and even if it doesn't
> > occur in any known language, doesn't necessarily mean that it is
> > unnaturalistic (while on the other hand, if it occurs in just
> > *one* natlang, that is sufficient to award it with the hallmark
> > of naturalism).
> These seem to me to be different phrasings of what I said - having an
> occasional rare feature is common, unattested != necessarily
> unnatural, and attestedness defines at least the outside fringes of
> what is 'natural but unusual'.

Yes.  As I said before, to give an example from my main conlang,
I know of no natlang precedent for the degree-of-volition marking
system of Old Albic, but I have seen things that get close, and
I feel that there is nothing unnatural about it - it may be very
rare, but it is certainly possible.

> > However, "exotic"
> > features should not be overused - if you combine 20 features
> > each of which occurs in less than 1% of the world's languages,
> > you are probably heading in the wrong direction.
> I'd phrase this more neutrally, and say simply that if you do so, and
> those 1%s are mutually independent (rather than clustering), then your
> result is extremely unusual / weird for a natural language. If that's
> your thing, fine, if not, change it. :-P

Sure.  There is nothing wrong _per se_ with "kitchen sink" conlangs,
but one *should* be aware of what he's doing.  A language with 20
independent features which each occur in less than 1% of the world's
languages is not necessarily a bad conlang; indeed, it can be very 
interesting.  It is just that I personally prefer keeping "weirdness"
to a more moderate level.

> > * taxonomic vocabulary (as in the 17th-century "philosophical"
> >  languages)
> One has to caveat this one to say that there are frequently clusters
> of words that are similar both semantically and formally, as well as
> things like triconsonantal root systems which have forms that are
> more-or-less regular transformations of those root meanings.
> But a true multilevel taxonomy is a different thing.

Certainly, many, if not all, languages have word families where dozens
of semantically related words are derived from the same root, and this
happens *a lot* in Semitic languages, for instance.  Yet, taxonomic
conlangs such as Dalgarno's or Wilkins' go much further and try to
abolish arbitrariness in word formation by deriving all words from a
single universal taxonomy of idea, and this is not without problems.
At any rate, I would say that it is unnatural.

> > * stack-based syntax (as in Fith)
> > * implementation of a scheme of formal logic
> Aww. :'-(
> :-P

Well, they *work*, but I wouldn't expect any of these in a human

> > There are probably many more, not to mention languages of
> > non-human beings that lie entirely outside human possibilities
> > of articulation or perception (e.g., ultrasonic languages).
> This of course brings up the question of what makes for a naturalistic
> language for non-humans.
> Do we assume the same psychological-ish universals? The same things
> re. economies of features (as you mentioned above)? Anything?

The problem is that we have no data.  We know only *one* language-
using species (our own), and thus simply do not know what languages
of other sapient species may look like.  We can only speculate that
they would be similar to ours in certain regards (such as having
phonologies that make economic use of phonetic features, and all that).
When designing a bizarre non-human conlang, one can indeed always say,
"It may be weird for us, but it is perfectly natural to them", and
there is no way to prove that wrong.

> > As I have said before, if, just for an example, the Afroasiatic
> > family had died out without trace (or evolved in a very different
> > direction), most linguists would consider a morphology in which
> > the lexical meaning of a word is encoded in three consonants
> > between vowels are infixed to make concrete word-forms at least
> > very weird, if not unnatural.  But we know that it happened in
> > Afroasiatic, so we know that it is not unnatural.
> Indeed. And Alex gave me the example that clicks are rare enough that,
> had they not happened to exist, we might consider unnatural.

Yes.  Clicks are indeed rather weird, and if there weren't languages
using them in Africa, we'd think they could never occur in natlangs.

> > Second, there are features that are prescribed by the limits of
> > what humans can produce, perceive and process.  For instance, no
> > human language uses sounds humans cannot hear ;)
> Except for those that deaf people can't hear, and people with cleft
> palates can't produce well. ;-)

Of course.  I was thinking of "normally" developed humans without
such conditions.

> One could easily imagine (or even cite) cases where local populations
> of humans happen to have such a variation in high enough concentration
> that the language they use as a group compensates for it. (Martha's
> Vineyard is canonical here.)

I am not sure what you mean.  Do you mean a population where a
specific disorder or inability is so widespread (e.g., due to
inbreeding) that they change their L1 to accomodate for that?
This would indeed be imaginable.

> > Finally, there are things which *could* occur in human languages
> > but *do not*, for whichever reason.  It is likely that many of
> > the universals discussed by linguistis fall into this category.
> Which ones? To me, this is the interesting category, because it's the
> set of things to try to do. ;-)

I don't know.  I am not a professional language typologist or
language psychologist.  Indeed, the "unattested but possible"
category is of particular interest to conlangers, both naturalistic
(exploring unattested combinations of attested features, evolving
languages which could have evolved but didn't [I am thinking of
conlangs such as Brithenig here]) and non-naturalistic (trying out
radical departures from natlangs, such as logical languages or

> > I would always be careful with such adjectives as "ideal" or
> > "optimal".  There are usually many ways to meet the design
> > criteria one sets for a project, and it is hard to say which
> > solution is the best.  As you say, you usually get trade-offs
> > between conflicting goals.
> Correct. Which is why it is non-exclusive (there will be many 'ideal'
> or 'optimal' languages for any given valuation, due simply to the
> large number of choices that are entirely arbitrary and
> value-neutral), as well as valuation-dependent (one can't evaluate
> which is better without *first* determining where to set that
> trade-off preference).

Yes.  As I said several times, each conlang ought to be gauged
against its design goals (express or implied) set by the author.
This should be, I think, the very first rule of conlang critique.

... brought to you by the Weeping Elf