Re: Stack-based syntax (was: affixes)

Hallo!

On Tue, 22 Feb 2005 18:38:22 +0000,
Ray Brown <ray.brown@...> wrote:

> On Monday, February 21, 2005, at 08:12 , Jörg Rhiemeier wrote:
>
> > Hallo!
> >
> > On Sun, 20 Feb 2005 19:09:57 +0000,
> > Ray Brown <ray.brown@...> wrote:
> >
> >> On Saturday, February 19, 2005, at 11:01 , # 1 wrote:
> [snip]
> >> I am strongly of the opinion that if one is going to construct a truly
> >> stack-based syntax one simply has to think in terms quite different from
> >> traditional western 'parts of speech'.
> >
> > Maybe.  Fith still has nouns and verbs, though.  But at least,
> > adjectives and adverbs are one and the same in Fith.  The difference
> > is merely what is on top the stack, an NP or a clause.
>
> Yes, I have now looked at Fith. I was writing in general terms about how I
> perceived a stack-based syntax.
Yes.

> Fith does indeed have a part of speech called a 'modifier' which acts as
> an adjective if the top of the stack is a NP or as an adverb if it is a VP.
>   The modifier combines both lexical meaning and also acts as an operator.
> I assume this must have some parallel in FORTH.
>
> I am well acquainted with stacks and their various uses. Probably because
> I associate 'stack-based syntax' or LIFO syntax with the evaluation of
> 'Reverse Polish' expressions, I would prefer operators and lexical items
> to be kept distinct.
Comparisons between human languages (or languages of some other sort
of sapient beings; we definitely go beyond *human* languages when
discussing stack-based languages) and computer programming languages
are always problematic.  I have heard of a grammar of Vietnamese
which describes the language in terms of "classes" and "methods";
obviously done by an object-oriented programmer on the rampage ;-)

> >>> Articles would also work that way to tell the definiteness of the noun
> >>
> >> _Many_ languages do not have articles. I believe in fact the majority do
> >> not. So I think one needs to ask whether they would figure at all.
> >
> > Yep.  I think Fith has articles, though, which are indeed unary
> > operators.
>
> It does indeed & they are unary operators - but their use is optional,
> which is not the case with unary oprators like NEG or NOT.
Yep.

> >>> Case marks would act that way with a single noun
> >>
> >> Surely not! Is not the whole point of case markings that they show how
> >> the
> >> noun/pronoun _relates_ to some other part of the sentence (usually the
> >> verb). As I see it, the the case markings are binary operators.
> >
> > Yes.  Case markers in Fith are binary operators: they take the NP
> > on top of the stack and the item (phrase or clause) below it, and
> > knit them together.
>
> Which is exactly what one expects. Although in the version of Fith I
> downloaded, it says "Nouns are not marked for number, gender or case.." I
> assume we're talking here of postpositions.
Yes.  I was referring to the items Jeffrey calls postpositions.

> >>> And the intransitive verbs also because they only affect one noun
> >>
> >> I do not see how a lexical category like verb should act as an operator.
> >> Isn't it something more like: singing, John, NOM = John is singing?
> >
> > It is a question of semantics.  In Fith, verbs are indeed operators.
> > Intransitive verbs are unary operators, transitive verbs are binary
> > operators.
>
> You're right. Once again the Fith verb is combining both lexical meaning
> and operator.
Yes.

> >  Perhaps more "part-of-speech" thinking involved here
> > than there should be.
>
> I agree entirely. My impression is that the Fithians, tho intelligent
> marsupials, are not so very alien from us humans. IMO the language is too
> dependent on western IE language structures. It is even necessary,
> apparently, for the Fithians to use hand signals in conjunction with
> speech to clarify parts of speech - "The exact part of speech is marked by
> a hand signal..."
Yes.  There is much "SAE" thinking involved in the grammar of Fith,
which is misfortunate.  And the hand gestures associated with the
language are a device which strikes me as hackish and clumsy.
It seems as if we could do better.  Nevertheless, Fith stands out
as a language that breaks out of the corset of human language thinking
at least in one regard.  I have seen "alien" languages that don't
look more alien than, say, Old Albic.

> If I were asked to construct a language with stack-based syntax I would
> keep lexical items strictly separate from operators. The lexical items
> would be my 'literals' and the idea of 'part of speech' has no meaning in
> that context.
Yes, that would be nicer.

> >>> Other operators will simply link two arguments in different ways like
> >>> conjonctions and other in more sophisticated ways like transitive verbs.
> >>
> >> I would see a transitive verb more like: (John NOM (loving ACC Jennifer)
> >> )
> >> which in postifix (stack-based) form would be: John, loving, Jennifer,
> >> ACC,
> >>   NOM.
> >
> > Yes, that's more elegant.  Keeping content words and operators nicely
> > apart.
>
> Yes, certainly more what I am used to.
Yep.

> >> [...]
> >>
> >>> On Sunday, February 20, 2005, at 03:56 , # 1 wrote:
> >> [snip]
> >>
> >>> Isn't exactly as an SOV, postpositionnal, noun-adj language?
> >>
> >> IMO no - not exactly. Certainly such a language might be a good place to
> >> start.
> >
> > A stack-based language is not only "not exactly" an "SOV,
> > postpositional,
> > noun-adj language", it is *not at all*.
>
> I agree - I was indulging in typical British litotes,
Of course.  I am aware of the fact that _not exactly_ sometimes
takes on the meaning `not at all'.

>       and I didn't want to
> be too discouraging to Max. But you are right - it ain't a SOV at all.
Yes.  The term "SOV" is part of a typology meant for the kind of
languages humans use, but stack-based languages, even if they happen
to have "nouns" and "verbs" like Fith, operate outside that frame,
so a resemblance of a stack-based language to an SOV etc. language
can only be superficial.

> > It lies wholly outside the
> > range of human language structures.  A simple clause in a stack-based
> > language may perhaps look like one in an SOV, postpositional, noun-adj
> > language (as a simple clause in Fith indeed looks like), but that
> > resemblance is merely superficial, because the clause is parsed in an
> > utterly different way from *any* human language.
>
> I agree entirely. Unfortunately IMO because Fith is described in terms of
> the familiar Latinate 'parts of speech' one can get the impression that
> the stack is merely another way of presenting a SOV human language.
Well, I think that even in the concrete case of Fith, it is clear that
it isn't.  But the terms "noun", "verb", "postposition" etc. are not
very well chosen.

>       In
> fact, as you say, a true stack-based language will be utterly different
> from any human language. But it is not easy for us to think in alien terms
>   :)
Right.  And I expect languages of real alien intelligences to be much
more bizarre than any science-fictional speculations that have been
made so far.  We know no non-human sapients and no non-human sapients'
languages, so our models of alien languages are inevitably
anthropocentric.

> [snip]
> >>>
> >>> A sentence like:
> >>>
> >>> dog the big cat your small love = the big dog loves your small cat
> >>
> >> But this doesn't clearly separate lexical items and operators. This would
> >> imply, for example, that 'love' is a combination of lexical item and
> >> operator. IMO in a truly stack-based system, lexical items and operators
> >> should be kept distincr.
> >
> > That would at least be more elegant.  However, in Fith, verbs are
> > operators as well as lexical items, as in the example sentence above.
>
> True - as I said, I wrote the remarks above before looking again at Fith.
> To me Fith seems a bit of a compromise, but maybe this true of FORTH - I
> don't know.
It is long ago (almost 20 years) since I had my last exposure to FORTH,
and don't remember much of it.  So I cannot comment on this.

> [snip]
> >> IMO with a stack-based syntax, lexical items are literals and operators
> >> make explicit how the literals relate to one another.
> >
> > Yes, that would be indeed much more elegant than Jeffrey's Fith.
>
> Thanks  :)
Greetings,

Jörg.

Reply