Hallo conlangers!

On Wed, 18 Apr 2012 10:17:01 +0200 Christophe Grandsire-Koevoets wrote:

> On 17 April 2012 17:23, George Corley <[log in to unmask]> wrote:
> 
> > >
> > > The notion that natlangs are probably not LIFO is a Conlang myth. On the
> > > contrary, natlangs probably are LIFO. So achieving fluency in a LIFO
> > > conlang would surprise only those who subscribe to this Conlang myth.
> > >
> > > The weirdnesses of Fith are to be found among the stack operators, which
> > > are documented online.

What remains if one takes them away is indeed something not very
dissimilar from human languages - Shallow Fith.

> > > I myself wouldn't claim that those are completely
> > > unusable, but I would claim that their usability correlates with the
> > > brainpower (i.e. mental brute-strength) of the user -- which is not true
> > > for ordinary natlang processing.
> > >
> > > --And.
> > >
> >
> > How, exactly, could natlangs be LIFO?
> >
> 
> After reading some more on that issue, my understanding is that if natlangs
> are *strictly* context-free, then they can be parsed by a machine equipped
> with no more than a LIFO stack (it's a mathematical truth).

Yep.  Though the class that can be parsed by a *deterministic*
machine with no more than a LIFO stack is indeed smaller than
the class of context-free languages.  Or so I remember from the
formal language theory lessons I took when studying computer
science, but that was 20 years ago.

>       It doesn't mean
> that the human brain itself parses them that way though (there's not a
> single piece of evidence that it does, and plenty of indirect evidence that
> it *may* not), just that a LIFO stack would be *enough* to do so. So while,
> under that condition, And's claim may indeed be correct (that natlangs are
> LIFO), it doesn't say anything about how humans *actually* parse natlangs

Indeed not.  Even if all natlangs were members of DPDL (the class
of languages that can be parsed by deterministic LIFO machines),
that wouldn't mean that humans use such a machine to parse natlangs.

> (And's mistake is to confuse sufficiency with necessity. If there's
> something my biology and physiology classes have taught me -I took them at
> university level, it was a mandatory part of my curriculum-, it's that
> biological systems are *messy*, are almost never exactly fit for purpose,
> and are often redundant and overkill compared to what they are supposed to
> do. There's no reason to expect our language ability to be otherwise).

Right; biological systems almost never are "optimal" - they are
just *good enough* to survive, and in many cases, biological
systems serving a particular purpose evolved from something that
served (or still serves) some other purpose.  Language certainly
is such a case!

It seems to me as if And's chief mistake was to confuse minds and
machines ;)

> The problem is, there's quite a lot of evidence around that shows that
> natlangs are *not* strictly context-free. The case of Swiss German, for
> example, is difficult to deny (those who tried always resort to arguments
> that are equivalent to saying that the native speakers are just wrong,
> which to me sounds like prescriptivism gone wild), and led to the
> definition of the so-called "mildly context-sensitive languages", which
> need a machine with at least a stack of *stacks* to be parsed successfully.
> And now it seems that Suffixaufnahme may be another case that doesn't even
> fall under those mildly context-sensitive languages.

Maybe.  I have my doubts that natlangs are always strictly context-
free, too.  My lessons in formal language theory are long past, and
I have forgotten much of them, but isn't anything where three
elements have to agree with each other beyond what is possible with
context-free grammars?  At least, I have seen many rules in formal
descriptions of natlangs that were of the form

X -> P /Q_R (X: nonterminal symbol; P,Q,R: any strings of symbols)

which is the basic pattern of a context-*sensitive* rule.  In
a context-*free* language, a grammar can be given where Q and
R are always zero.  Hence, I guess that natlangs are usually
context-sensitive, not context-free.

> My own opinion is that trying to describe language using a strict
> mathematical framework is bound to fail, because language isn't a
> mathematical phenomenon, but a socio-biological one. As I wrote above, such
> phenomena are "messy", they don't have strict limits that can be captured
> by a theoretical framework. They can be *approached* with one, but never be
> *exactly* described by it (which doesn't mean that it's wrong to try :) ).

Yes.  Describing a natlang with mathematical formulae is much like
describing the shape of a living being with basic geometric forms.
It won't get you far.

> Moreover, the whole Chomsky hierarchy is based on Turing machines and less
> able automatons, and all those parsing mechanisms have one thing in common:
> they can't change their own programming.

Indeed not - but AFAIK it has been proven that even a self-
reprogramming machine can't be more powerful than a Turing machine
(but don't ask me for sources).

>       Our brain has one defining quality
> that defeats any attempts to compare it to such machines: it is
> self-modifying, and its ability for self-modification is *central* to how
> it functions! So any theory that doesn't take this ability for
> self-modification in mind will never correctly describe how the brains
> functions, whatever part of its activities one is looking at.

Don't ask me how powerful self-modifying machines are, but I am
pretty sure that they don't exceed Turing machines (see above)
- but those still can do much more than LIFO machines!

> Personally, I don't know whether the brains actually has an actual language
> parsing facility, and even if it does it might not the central part of its
> language facility, more a last resort thing, and mostly conscious rather
> than subconscious.

It would not surprise me if there was *no* neural circuitry in
the brain whose *exclusive* function is processing language.
It may be the same as with the articulatory organs, which all
also serve non-linguistic functions, and have been there with
those functions long, long before anything remotely recognizable
as Language happened to appear. 

> [...]
> 
> All this doesn't mean that we may not be able to describe our language
> ability with a framework based on a mathematical parsing facility like
> those used by the Chomsky hierarchy (and, although there's some evidence
> that seems to point otherwise, maybe natlangs are indeed context-free and
> thus only need a LIFO stack for successful parsing), just that such a
> framework doesn't necessarily tell us anything about *how we humans do it*.

Indeed, it won't tell us.  Such a theory would be useful for
developing computer software for processing natlangs, but it
would not tell us much about how *we* do it.

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