Re: [lojban] semantic parser - tersmu-0.1rc1

[John E Clifford <kali9putra@yahoo.com>]

----- Original Message ----
From: Jorge Llambías <jjllambias@gmail.com>
To: lojban@googlegroups.com
Sent: Sun, December 11, 2011 12:42:03 PM
Subject: Re: [lojban] semantic parser - tersmu-0.1rc1

On Sat, Dec 10, 2011 at 6:58 PM, John E Clifford <kali9putra@yahoo.com> wrote:
> Well, I am glad you know that quantifiers don't refer, since then you can
> understand my discomfort at seeing them treated as though they do: like names,
> in fact.

Quantifier terms are treated by the Lojban grammar just like names,
yes, even though they have different functions, names refer and
quantifier terms quantify bridi.

That is just what I am uncomfortable about (and have been -- unconsciously -- 
for decades)/

> To be sure, at a certain point (prenexing usually), they receive some
> different treatment, as they pass through negations and the like (or don't, as
> the case need be).

They receive different treatment at the point of interpretation. Names
can be moved to the prenex too, but the rules by which quantifier
terms can be moved to the prenex are more strict than for names. The
only reason to have these rules at all is that once they are in the
prenex the interpretation is plain, as it corresponds directly to
standard notation.

As I said, they are treated differently.  And the interesting question is, how 
do we know that what we get when we get everything prenex by these rules is what 
was intended in the original?  At the least, there should be a process of 
converting FOL into Lojban that would justify the reverse reconstruction.  As it 
is, we have only success and, perhaps, an intuition that this is the right way.

> But, while they don't refer, they do have a set of true
> instances and these are often dependent upon quantifiers in whose scope they
> lie.

That statement is problematic. It may be understandable when the
quantifier term binds the last free variable of a bridi, but it
doesn't really make sense when the quantifier term leaves some
variable still unbound. P(x) has a set of true instances for "x", and
the quantifier Ex: tells us that the cardinality of that set is at
least one. But Ex:P(x,y), with y still unbound, has no set of true
instances for x, since it is not a proposition. It has a (possibly
different) set of true instances for x for each different value of y.
So for y=a we have one set of true instances of x, for y=b another
set, and so on. The only meaningful set of true instances of x that
you can get for "Ex:P(x,y)" is the one that consists of the union of
all the sets that you get for each possible value of y. And that set
could even be empty, if ~Ey: Ex:(Px,y) happened to be true. So "Ex:"
tells us nothing about the cardinality of the union set.

Sorry, I assumed we were talking about expressions in sentences, not just 
floating around.  There are no free variables in Lojban sentences, so this 
remark would not apply in those cases -- nor would I mention satisfaction sets 
in the other cases.

> Thus, the {su'o plise} of your example appears to lie in the scope of two
> universals (I'm ignoring the negation for the moment, since it doesn't affect
> the issue) and to have the corresponding true instances.

It's in the scope of two universals, plus the negation you are
ignoring, plus a connective, and what's more important, it binds a
formula with two free variables, which means that it gives no
information about the cardinality of any single set.

Sorry, I don't see the other variable, unless it is plise2 (a cultivar?) which, 
if an invisible variable, is, of course, bound by a corresponding invisible 
quantifier.

> But the intention is
> that it have two instance sets, one for each quantifier: the hybrid set is
> neither of these (probably).

Whose intention is that?

The intention of the speaker, if that is what the final expansion to FOL 
produces (and, if it isn't, isn't the procedure a failure?).

> So this makes the sentence seem strange.  Your
> claim is that the sentence ought not be taken at face value, but understood as
> the result of applying a fairly simple (so far, anyhow) set of interpretation
> rules.

I don't think I made such claim, mainly because I don't understand
what you mean by "taking the sentence at face value". If I had to make
a claim, I would say that it ought indeed be taken at face value, and
I explained what that face value meant to me. If by "face value" you
mean "what it means in standard notation", then the sentence has no
face value at all since it involves non-standard notation, it is
uninterpretable as standard notation, and thus it cannot possibly be
taken at face value. If you see some other face value I'm not seeing,
what is it?

Well, to take the simplest example, at face value, {su'o plise} is in the scope 
two quantifiers, and ot is a simple sentence with a compound subject, so neither 
a compound sentence (an arbitrary ruling, I admit) nor a complex one, whereas, 
eventually, it will prove to be both and {su'o plise} will occur twice. each in 
the scope of only a single quantifier.

> And (so far, at least) these rules do regularly yield the right
> results.  But then the question is, how else are these rules justified?

What kind of justification are you after? If it's something involving
a set of true instances, you need to make that notion more explicit,
since a bridi with more than one variable does not have a well defined
single set of true instances for one of the variables.

Not sure how this applies, since the case does not seem to arise.  As you say, 
which apples will be in that set depend on the value of both the universals and 
the resulting set appears to be not what was wanted in either direction or, at 
best, the one's for the first case (boys) but not for the second (girls).

> They
> seem to have no basis in logic, probably because they deal with structures 
>logic
> does not allow,

That's unavoidable, isn't it?

Probably, if we are to have a speakable language.  But are the things we have 
really representing the underlying logic?

>intermediate constructions which are not to be taken too
> seriously (except that they usually are also sentences of Lojban) on the way 
to
> the final results.

Which intermediate construction is not to be taken too seriously? If
there was such intermediate construction, the whole thing falls apart.
The whole point of the rules is that there be no gaps in moving from
the non-standard form to the standard one.

?? "such intermediate construction" I suppose means "one that is not a Lojban 
sentence".  Well, is there a guarantee that the rules will never lead to one on 
the way to a correct interpretation?

> I can hope that eventually a Montague grammar will come
> along to justify moves that accomplish the same results in a rationalized way.
> Until then, success is probably good enough  -- and may always be.  But it 
does
> not present much in the way of guidelines when difficulties arise.

It's not so complicated. The only two reasonable options for "ko'a .e
ko'e broda su'o da" are:

(1) ge su'o da zo'u ko'a da broda gi su'o da zo'u ko'e da broda

(2) su'o da zo'u ge ko'a da broda gi ko'e da broda

The choice between the two is ultimately arbitrary, purely a matter of
convention. You can't expect the compact form to have some intrinsic
"face value" that tells you which one of them it is, it's merely a
matter of knowing whether ".e" has scope over "su'o" or viceversa. In
a natlang, we just live with the ambiguity. In Lojban, we don't want
that kind of ambiguity and so, by convention, decide that the compact
form corresponds to one and not the other. The only justification we
may add for the choice is which one makes the whole system nicer, and
in my view that means choosing (2) simply because I think it's better
to stick with left over right scope. What possible justification
beyond that are you looking for?

I assume you mean (1) here, since that is what your convention selects.  2 makes 
a bit more sense within the usual sort of construction rules, which allow 
collapse only for identical terms and so lead, as noted earlier, to giving {su'o 
da} broader scope.  But that is a trickier rule to apply, in either direction.

> I am not sure I understand your objection to the traditional solution to 
donkey
> sentences.  Do you know of cases where it just doesn't work?

Changing a narrow scope existential into a wide scope universal? And
already mentioned cases where that doesn't work: "most farmers who own
a donkey, beat it"

Sorry, why doesn't it work?  For all donkeys x, for most farmers y who own x, y 
beats x.

> The non-quantifier
> case is handled nicely by analogy: "Each of Mary and Jane is such that those 
>who
> love her want to marry her."

That's why I brought up the example, to support that connectives
should be treated the same way as quantifiers.

But this is a quantifier case from "one of Mary and Jane" to "each of Mary and 
Jane" (That being said, of course, there are viable analogies between 
quantifiers and sentential connectives -- not ones that can be always relied on, 
however).

mu'o mi'e xorxes

-- 
You received this message because you are subscribed to the Google Groups 
"lojban" group.
To post to this group, send email to lojban@googlegroups.com.
To unsubscribe from this group, send email to 
lojban+unsubscribe@googlegroups.com.
For more options, visit this group at 
http://groups.google.com/group/lojban?hl=en.

-- 
You received this message because you are subscribed to the Google Groups "lojban" group.
To post to this group, send email to lojban@googlegroups.com.
To unsubscribe from this group, send email to lojban+unsubscribe@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/lojban?hl=en.