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

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

Curse you, auto- correct!  "helter skelter" ( as I suspect you know, but "helper smelter" does sound weirdly appropriate)

Sent from my iPad

On Dec 9, 2011, at 9:24 AM, Jorge Llambías <jjllambias@gmail.com> wrote:

> On Thu, Dec 8, 2011 at 8:39 PM, John E. Clifford <kali9putra@yahoo.com> wrote:
>> I suppose my point is just that, when we make all these claims about being logical ( even in the narrow sense), one expects that the way that sentences are constructed has some rational connection to the logic below, not just helper smelter rules which happen to work, in some practical sense.  So, what I sketched briefly was one case of trying to make such a rule, where the sentences clearly meant the same thing from beginning to end, not just ending up right.  As I  say later, such rules may not be possible, but they do deserve a look.
> 
> The sentences have to mean the same thing from beginning to end with
> the helper smelter rules too, not just end up right. If there is one
> step where they mean something different, the helper smelter rule
> fails.
> 
> It seems that what bothers you is that Lojban has eks and giheks, and
> quantifiers and negations admitted as pseudo-arguments. You may want
> to call that illogical, and in some sense it is, but it's been part of
> Lojban and of Loglan from the beginning, so you can't be finding out
> about it just now.
> 
> Since these constructions are not part of standard notation, there
> have to be helper smelter rules to show how they come about from
> standard notation, or how to change them into standard notation.
> 
> I don't see anything particularly irrational in the helper smelter
> rules as I proposed them, and you didn't give enough information about
> any alternative rule you have in mind that you claim would make more
> rational connections, so I can't compare.
> 
>> As for what is ignored, the list in this case is fairly short: binding, instantiation, and role.  The one particular quantifier is governed by two universals ( more or less), which means it's instantiation has to take both into account,
> 
> How is "ro lo re prenu cu citka su'o plise" different from "la .djan.
> .e la .meris. citka su'o plise"? "Each of the two people ate an apple"
> vs. "Both John and Mary ate an apple"?
> 
One is complex, the other compound; one has a quantifier in the scope of another, the other doesn't (though the corresponding Skolem function would have to take both names into account, which may be a similar constraint); one has instances, the other doesn't.

> Or even better: How is "no lo re prenu cu citka su'o plise" different
> from "la .djan. na .e nai la .meris. citka su'o plise"? "None of the
> two people ate an apple" vs. "Neither John nor Mary ate an apple".
> 
> Assuming we agree those are in some sense equivalent (but who knows if
> I can make that assumption), extend it now to "Each of the two boys
> and one of the two girls ate an apple" vs "John and Paul, and either
> Mary or Alice but not both, ate an apple". Is there any reason for
> that not to be the reading of "ro lo re nanla .e pa lo re nixli cu
> citka su'o plise"?

Well, the equivalence depends on the assumption that the groups named have indeed just the members named, but that is rather trivial.  And, of course, the names version is hardly a reading of the quantifier version, though it is an instance of it, under certain further conventions.  Of course, shifting from quantifiers over variables to quantifiers on plural terms also changes the game somewhat, since we then have to deal with either set their of some sort or plural reference, though it is not clear that this expansion presents any real problems.
> 
>> while in the final analysis, the two particular quantifiers are each governed by a single quantifier and needs only to take that into account.  Neither of the two ultimate instantiations would be the one the single case would give,  and, while you may say that the quantifier expressions are not terms, they are certainly treated as such
> 
> So that's what bothers you, right? That a quantifier is treated
> syntactically as a term, when it is obviously not a term semantically.
> If that's what you wanted to point out as irrational, then I agree,
> there is some irrationality in that. But it's something that has
> always been part of Lojban.

Well, yes, I suppose it is.  Thank you.  I have maintained since 1976 (with occasional lapses) that there was something inherently wrong with the propaganda about Logjam and just now I see the crux of it: scope lines are scrambled and different quantifiers are treated as the same.  You cannot, for example, find veridical instance of the recto particular quantifiers in the examples with names.
> 
>> (the claim that that they can't shift positions doesn't enter here, nor, I would think, in analysis generally);they are, after all, joined just like names, not the overarching structural elements the are ( ultimately).
> 
> Right, and that's why we need helper smelter rules to sort that out.
> Or do you think there is some other way?
> 
>> Yes, every particular quantifier ( and universal, for that matter) is the same in function, but surely not in content, a {su'o plise} in one sentence points to different apples from that in another sentence. And similarly for a universal, if we are restricting our domain to just the immediately relevant groups, as seems to be the (unacknowledged) case here.
> 
> Is that also what bothers you, that a predicate may have different
> extensions in different instances of use due to domain of discourse
> shifts? This is a separate issue, and then you should also be worried
> about "la .djan. .e la .pol. cu prami la .meris." expanding as "la
> .djan. cu prami la .meris. .i je la .pol. cu prami la .meris.", since
> we can conceive of contexts in which "la .meris." has different
> referents in the expanded form but not in the collapsed form. The
> helper smelter rules assume a fixed domain of discourse so that "su'o
> plise" (or "ro plise", or "no plise") keeps the same domain every
> time.
> 

Not the domain, the actual instances.  We would reject a collapse of the second sentence here into the first if the two Marys were different, so why allow it when the two {su'o plise} point to different apples, even if from the same pile? 

>>>> (note the negation sign has here
>>>> been reassigned as part of the quantifier)
>>> 
>> The objection is, of course, to burying a negation ( another structural element) in a term, a non-structural element.
> 
> But that's what Lojban does and has always done. Surely you are not
> finding out about this just now.
> 
No, just finding out about the source of an abiding gut feeling, which I had never quite found the right words to express.

>>> They are two instances of one and the
>>> same sentential operator, just like two instances of "na ku" would be
>>> two instances of one and the same sentential operator. (They are
>>> "terms" only in the sense that Lojban's formal grammar calls them
>>> that, but that's just Lojban being sloppy with technical terms as
>>> usual.)
>>> 
>> Except that quantifiers, unlike negations, introduce references to things.
> 
> In the sense that they need a domain of quantification? Yes. And the
> domain of quantification should be maintained for quantifier "terms"
> that are shared by eks or giheks.

Not domain, instances are the problem.

> 
>>> That involves a donkey pronoun, and we don't as yet have spelled out
>>> proper rules to handle them. This is not part of the kind of unpacking
>>> rules we were discussing.
>>> 
>> But it is just the kind of rule that needs to be discussed.
> 
> Sure. But let's not confuse it with the helper smelter rules that deal
> with the much more tractable issue of expanding eks and giheks and
> moving "quantifier terms" and negations to the prenex, and the
> slightly more hairy but still manageable issue of dealing with
> implicit binding of variables. The jury is still out on donkey
> anaphora, but eks, giheks, quantifier terms and negations only need a
> reasonable convention. Let's not muddle the simple issues with the
> truly complicated ones.
> 
Well, of course, I don't think donkey sentences are problems; we know exactly how to handle them and always have.  they do present problems for rules of your sort which try to take items one at a time and in order, rather than a more global (at least context-sensitive) approach.  As soon as you assign a {su'o} to all "floating" variables, you have eliminated the possibility of donkey sentences, which may or may not cripple your results, but seems rather arbitrary, given how natural languages work.

Again, this is a theoretical ( even aesthetic) comment.  The needs of a speakable language override logical clarity.  If the crude rules always give the right results, what's the problem? Even a little braggadocio does no harm.  But a bit of concern with justifying the rules by something other than their results wouldn't hurt either.
> 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.