On 22/04/2012 22:44, And Rosta wrote:
> R A Brown, On 22/04/2012 20:26:
[snip]
>>
>> If the expression is read sequentially in infix
>> notation, i.e. 1 + 2 × 3 - 4 ÷ 5, then to achieve the
>> correct result you will indeed need to construct a
>> parse tree. But if the _postfix_ version, i.e. 1 2 3 ×
>>  + 4 5 ÷ - is read sequentially, it is easily evaluated
>>  using a only stack; to construct a parse tree would be
>>  utterly pointless.
>
> Let me see if I understand you, and be patient if I
> don't:
>
> Are you saying that a tree structure of operators and
> operands is pointless because the parser evaluates as it
>  goes. E.g. in, say, RPN "1 2 +", one the parser has
> combined "+" with "1" and "2", it outputs or places on
> the stack not the single tree "1,2,+", but rather the
> value "3"?

Yes -  1 gets pushed on to the stack, then 2 gets pushed on.
We then meet the operator + which means "pop the top two
values off the stack, add them and push the answer back onto
to stack."  In the case + and * the order in which the
operation is carried out is immaterial, but in the case of -
and / then we have to subtract the top value from the next
top value or divide the next top value by the top value -
*but in all the operations no tree is needed*.

Obviously in the case of 1 2 + there is no obvious advantage
over 1 + 2.  The advantage of postfix notation becomes
clearer when the expression is longer and contains a mix of
different operators with different orders of precedence.
The example Jeffrey gives is 1 + 2 × 3 - 4 ÷ 5.

If we evaluate this simply by working from left to right we
would get:
(1 + 2) × 3 - 4 ÷ 5.
(3 × 3) - 4 ÷ 5
(9 - 4) ÷ 5
5 ÷ 5
1

The answer 1 would be marked incorrect by any maths teacher
and not what anyone who is mathematical savvy would evaluate
the expression to.  There is a convention that × and ÷ have
a higher precedence than + and -, i.e. 2 × 3 and 4 ÷ 5 must
be evaluated first, i.e.
1 + (2 × 3) - (4 ÷ 5)
(1 + 6) - 0.8
7 - 0.8
6.2

Also in infix notation the order of precedence can be (as
indeed my "workings out" in the first example shows) altered
by the use of brackets.

Clearly a computer cannot correctly evaluate 1 + 2 × 3 - 4 ÷
5 by simply working through from left to right; it has to
accommodate the human conventions of order of precedence
(and use of brackets).  A parse tree, in this case a simple
binary tree, is a convenient way of doing it (tho other
methods could be used).

> If this isn't what you're saying, then please try again
> to explain.

It is - tho in agreeing, I've expanded a little   :)

> If this is what you're saying, I don't understand what
> the linguistic equivalent of "3" in contrast to "1,2,+"
> is. That is, "1,2,+" is to "John, Mary, loves" as "3" is
>  to what? I have been supposing that parsing of both
> language in general and fith in particular works by
> reading in the symbols "1","2","+"
> ("John","Mary","loves") and outputting a tree structure.

I think the problem is, as I wrote in a later email
yesterday evening, that we (or perhaps just I) have been
confusing 'parsing' with 'evaluating'.

As I wrote in that second email:
{quote}
......... . the parse tree of both [infix and postfix]
expressions is identical - a binary tree.  If you do an
inorder traversal of the tree you read off the familiar
infix notation, if you do a post-order traversal you read
off the postfix notation.

But the point is that the parse tree may be a useful way for
a computer to read off and evaluate the infix notation (tho
not the only way), but it is utterly pointless in the case
of the postfix notational.
{quote}

The point is that _parsing_ a postfix expression inorder to
evaluate it is utterly pointless.  How this would apply to
language generally and not the narrow domain of mathematical
expressions is IMO more controversial and more of a
challenge to those who want to try it.

We generally understand language as we hear it or read it -
tho what goes on the mind is quite remarkable and AFAIK not
properly understood.  But many of us have properly met
occasions where some sort of analysis or parsing has been
necessary.  I recall many years ago when I was researching
for my M.Litt degree, coming across a sentence in a German
book that went on and on and on.  There was - and I don't
exaggerate - a full half-page of text before I met the
full-stop (period).  I was completely lost and the only way
I could make sense of the thing was to take the long period
apart and construct what, in effect, was a rough & ready
parse tree.

Now if I were a Fithian, that could never have occurred   :)

>
> In this discussion I had been abstracting away from the
> fact that the input to the parser is phonological (i.e.
> stuff that is phonetically interpreted) and the output is
> syntactic (i.e. stuff that is semantically interpreted).
> I mention this in case that abstraction created
> confusion, and you took me to be claiming that the
> eventual output of the parser is a mere structuring of
> phonological forms.

Not at all  - the input is phonological (or, as you say,
written form which represents the phonological input) and
the output is stuff that is semantically understood.  As far
as I see we are basically in agreement here.

But I deliberately avoided including the word 'syntactic'
because, I think, this is where the problem lies.  To move
from the input to the output some sort of syntactic parsing
or pattern matching or something) is needed  in the case of
human language.  How this is done is another matter and not
really relevant to this discussion - but we both agree that
parse trees are useful devices in demonstrating this.

> I assume that parsing of all languages, including Fith,
> takes a sequence of phonological forms as input and
> outputs a (semantically interpretable) syntactic tree.

If you are going to _parse_ Fith, then you can, if you wish,
out put a parse tree - but this should be a waste of time.
The point is that Fithians do not need any parsing (or
similar) process to go on in their minds as each understands
what the other says, i.e. as Fithian A speaks, Fithian B
receives the input and understands it as A speaks by simple
stack manipulation in its mind and by nothing else.

>> What Jeffrey was attempting to do was to produce a
>> "postfix language" - how successful he was or not, I
>> leave to others to judge. It was intended that, just as
>> postfix mathematical expressions can be evaluated using
>> nothing but a stack, so also would Fith need nothing
>> more than a stack - a parse tree would be quite
>> pointless.
>
> Okay. I don't see the distinction between an evaluation
> and a syntactic tree.

A syntactic tree is just that - a tree which demonstrates
the syntactic structure of the utterance.  It tells us, in
itself, nothing about what is _meant_.  It is true, such a
tree may help understanding a difficult or abstruse sentence.

Evaluation is to do with the meaning.  Exactly how the human
brain manipulates phonological input and gives us semantic
output is AFAIK not really understood - but I'm not up to
date with all the latest research.

But, as I understood Jeffrey, the Fith experiment was to
produce a language in which evaluation, i.e. comprehension
of meaning, would be possible by beings whose brain
supported only stacks as the reasoning process.

> It sounds as though you reckon the whole point of Fith
> was to make this distinction (that I don't apprehend). Is
> that right?

Yes.

>>> Setting aside Fith's stack conjunctions (which you
>>> consider incidental to Fith as a thought
>>> experiment),
>>
>> I consider them incidental in that Jeffrey's intention
>>  was to construct a language of postfix structure that
>>  needed *only* a stack for its evaluation (i.e. where a
>>  parse tree is redundant); the 'stack conjunctions'
>> were meant to be the analog of arithmetic operators in
>> the postfix expression 1 2 3 × + 4 5 ÷ -.
>
> That's not how I have understood it.

Again a difference in our understanding of what Jeffrey
intended - which is why I wish he were still around to
explain things.

> Take the sentence "zhong hong non zhong hong lo krai" =
> "man nation without man nation with hates" = "Man without
> nation hates man with nation". There are no stack
> conjunctions in this, and I understood "non" to be an
> operator, with operands "zhong" and "hong", and "lo" to
> be an operator, with operands "zhong" and "hong", and
> "krai" to be an operator with operands "zhong hong non"
> and "zhong hong lo".

Yes - "zhong hong non zhong hong lo krai" is, in fact,
nothing more than an example of a SOV object language with
postpositions - not exactly uncommon among human languages!

If this is all that Fith was then, indeed, I would agree
that it is nothing remarkable. But the sentence you give,
altho grammatical Fith, is an example Jörg's 'Shallow Fith',
see:
http://www.langmaker.com/shallowfith.htm

Of it, Jörg wrote:
{quote}
Shallow Fith is an interlanguage humans and Fithians could
use to communicate with each other.  Basically, it is a
proper subset of Fith, with strict limits to stack depth and
with most of the stack operators removed.  The Shallow Fith
sentences could be parsed as SOV/postposition/noun-adjective
sentences by humans and as very simple LIFO sentences by
Fithians.

*We are leaving the ground of Jeffrey Henning's original
creation here*; this is my own invention .....
{unquote - emphasis is mine}

>>> Fith looks to me like it has RPN syntax in which a
>>> stack builds up a tree of predicate--argument
>>> structure.

That was not, as I understood it, Jeffrey's intention. I had
a good deal of email exchanges with Jeffrey at that time
(tho not much of it about Fith) and met him when I visited
Boston, so I hope I have not misunderstood him.

[snip]
>
> Before assessing the success of Jeffrey's experiment or
> challenging Miles, we need to get a better collective
> handle on what the essence of the experiment was.
>
> You understand the essence to be its treelessness,
> right?

I think that is probably misleading.  As I have observed
above, the parse trees of the infix expression 1 + 2 × 3 - 4
÷ 5 and the postfix expression 1 2 3 × + 4 5 ÷ - is,
unsurprisingly, the same.  One can produce trees to
represent Fith sentences, but such a process is redundant -
at least for Fithians.

The essence of Fith, as I understand it, is that Fithians
evaluate/ comprehend meaning using nothing more complicated
than a stack structure - trees are quite redundant to the
process.  I imagine Fithians would find the whole concept of
trees and parsing, as we understand it, very difficult to
comprehend.

> But in what sense is the structure of "zhong hong non
> zhong hong lo krai" not the semantically interpretable
> tree rooted in "krai" (i.e. the syntactic counterpart of
>  the phonological form "krai") with dependents "non" and
>  "lo" and so on?

It is not and, as I have said, if Fith was nothing more than
SOV language with postpositions and post-posited adjective
it would be very unremarkable.

>>>>> Not really. A Fith parse still yields a tree (--
>>>>> sometimes one with crossing branches).
>>>>
>>>> Example(s) please.
>>>
>>> I think the Ordering Conjunctions will create
>>> crossing branches.
>>
>> If this true, then Jeffrey's thought experiment has
>> spectacularly failed. There certainly should be no
>> crossing branches.

I retract that statement.  What I will say is that if the
use of trees are _necessary_ in order to comprehend the
sentence, then Jeffrey's thought experiment has failed.

[snip]
>
> Take the example "John snores Mary knows" (= Mary knows
> John snores).

But is a relex of "John snores Mary knows" actually valid
Fith.  The language seems to me to be in too fragmentary a
state to give the Fith equivalent of "Mary knows John
snores" because "John snores" is a noun clause, being the
object of "John knows."  Many (most?) languages require tat
the be explicitly marked, i.e. Mary knows that John snores.

Some, like Latin, have a special construction for the object
clause; in the case it is the accusative and infinitive
construction, i.e.
Maria   scit  Iohannem stertere.
May.NOM knows John.ACC to-snore

We have no example of the handling of subordinate clauses in
Fith.

> "John Mary snores shen knows" means the same.

No it doesn't. On
http://www.langmaker.com/stackconj.htm
it is quite clearly stated that:
"shen ( n1 n2 - n2 n1 ) exchanges the stack positions of the
top two stack items"

I.e. "John Mary snores shen knows" is the equivalent of
"Mary John snores knows" not of "John snores Mary knows."

But whether any of these sentences are valid Fith (assuming
words are relexed) is IMO doubtful.

I think, however, we are talking at cross purposes.

As far as I understand it, parse trees (or any other trees)
are irrelevant to Fith *as used by Fithians.* I would expect
long Fith sentences to be incomprehensible to humans without
some use of such trees.

In summary:
- My understanding of Fith is that it is an attempt to
produce a language usable for beings who mental faculties
allow nothing other than a stack structure.
- that considerations of parse trees are not directly relevant.

As for the "stack conjunctions" on
http://www.langmaker.com/stackconj.htm

I think this something that anyone who wanted to experiment
with such a stack based language would do well to revisit.
Part of the problem, in my view, is that Jeffrey was trying
to build into the language means of flexibility and
stylistic variation.  In my view the language was not
sufficiently developed to do this at  this stage.  There
seems to me to be some ad-hoc-ness about this page.

Indeed, even on http://www.langmaker.com/fithpos.htm I think
we see something of the problem of creating such a language.
  Jeffrey wrote "The following analysis of the parts of
speech is heavily influenced by Western European languages;
this is not at all how a Fithian would describe the language."

Yep - any attempt to construct a purely stack-based language
is going to have the problem that we are creating a language
for being with a quite different mental structure than ours
- it ain't easy.  That to me is the challenge.

If there is a "LIFO myth" and a "Fith myth" it is that:
(a)a language could not be comprehended by an alien race
using LIFO structure alone, and the only use of a LIFO
structure is in the building up of parse trees.
- that Fith is nothing more than 'Shallow Fith' (i.e. a SOV
language with postpositions and post-posited adjectives)
with weird 'stack conjunctions.'

I maintain that
- there is nothing per_se that makes (a) impossible (tho
till humans meet such a race we cannot see definitely that
it is possible);
- that Fith was an attempt to do this, i.e its aim is to be
rather more than (b).

But IMO any attempt to imagine an alien (i.e. non-human)
form of thought is bound to be non-trivial - but might
produce something a little more convincing than the bog
standard sci-fi races who are boringly humanoid.

----------------------------------------------------------

On 16/04/2012 23:36, Miles Forster wrote:
> Right, thanks. I think I do understand the LIFO grammar.
>  However, I don't know whether the LIFO grammar makes it
> so that there can only be one way to do certain things
> like relative clauses or connectives.

That would be interesting to discover   ;)

> Also, if I am to make my own LIFO conlang, will people
> still believe it to be impossible to learn for humans?

Personally I think that any purely LIFO language *with any
depth* will be difficult, if not impossible, for humans to
use properly.  Some 'shallow' interlingua will be needed.
But, until the thing is properly done, we cannot be entirely
certain.

> I think I'd definitely try to keep as much of what Fith
> has to make sure I'm not "cheating". That means I'd keep
> the stack manipulation devices it has, but maybe change
> some or add some.

Personally, I think the 'stack conjunctions' do need revisiting.

> I'd also completely redo most of the phonology and
> dictionary, I assume.

Neither of those are essential to the project.  Indeed, it
never ceases to amaze me that alien phonologies are so
nearly humanoid.  It seems to me entirely probably that
intelligent beings have evolved somewhere in our vast
universe that have quite different sound tracts from any we
know.

> I'll have to ponder all this further.

Please do   :)

If you are still undaunted, I say go for it!

-- 
Ray
==================================
http://www.carolandray.plus.com
==================================
Nid rhy hen neb i ddysgu.
There's none too old to learn.
[WELSH PROVERB]