Re: Stack-based human languages? (was Re: My Three Assertions)

Hi!

Jörg Rhiemeier <joerg_rhiemeier@...> writes:
> > > Isn't that just a matter of staying within 7+-2 bounds? I don't see
> > > anything inherently more difficult about the *syntax* that makes it
> > > human-incompatible.
> >
> > I think so, too.  SOV and OSV syntax are just like Fithian.
Ok, this statement was too absolute. :-)

> I don't think so.
Ok.

> Fith allows a kind of stack acrobatic that goes well beyond anything
> possible in human languages, in fact going beyond what human minds
> can process in real time.  The resemblance of Fith to verb-final
> human languages is only superficial.
:-)

Yes, the stack operations are special and crucial for distinguishing
Fith from natlangs, that's true.  E.g. references to the nth stack
cell, duplication, swapping etc. are very different from how natlangs
work.

I just wanted to state that a stack is an appropriate data structure
for many structures in natural languages.  The resemblance is not
superficial, I think.  Only natural languages, in contrast to Fith, do
not allow direct operations on the stack and they use it to a limited
depth.

> >        A typical
> > *long* Japanese sentence (or a long German sentence, too) uses a stack
> > quite heavily.  It first pushes the nouns, then pops them with a verb,
> > pushes back the combined concepts, pushes more nouns, and finally pops
> > the rest.
>
> This is the first time I see any human language being described as
> stack-based.
Hehe. :-)

But you saw them described as rules like

   S -> NP VP

right?  And context-free grammars (which are not enough to describe
human language, of course -- just a small excerpt from some grammars)
are equivalent to stack automata.  Usually you start with a simple
approach like this to seethe real problems of natural language
processing.

> > Only a human language has a principle limitation in both the amount of
> > words a concept can be remembered when it is not used (=life time of a
> > stack cell) and the amount of data that can be remembered at the same
> > time (=stack depth).
>
> These "limitations" are of such an importance that they render the
> stack model of little use for describing human languages of any kind,
> even verb-final ones.
Reading my mail again I think I posed my point too radically.
I actually fully agree with you. :-)

> > ... daß Peter Maria Rolf  Essen  kochen     helfen     sehen      kann.
> >        push  push  push  push   poppoppush poppoppush poppoppush poppush
>
> This is a rather contrived example.
Yes. :)

> Few if any German L1 speakers talk that way.
Indeed.

> And while a Fithian would parse it that way, I doubt that any human
> would.  Human languages have tree-based structures with depth
> limitations, not stack-based ones.
Trees are the result of parsing a language using a stack-based
machine.  Fith can be parsed into such trees as well.  The additional
operations for stack manipulations to co-indexing etc. are different
from what is usually needed for natural language processing, but there
is a common basis.

Still no? :-)

**Henrik