Print

Print


On Aug 12, 2015 6:41 AM, "Patrik Austin" <[log in to unmask]> wrote:
>
> [Daniel:] > And since I think this version of the language can be used,
well, I
> still say you haven't been specific in exactly which semantics define
> FL2. I listed 3 options, but there are clearly more.
>
> The way I see it I've made something that seems to work as it gives a
functional parse tree, but why it works is a matter of theoretical
interpretation.

As Logan pointed out, the parse tree isn't the whole language. The four FL2
-type languages I've presented are VERY different from one another. I
played around with some examples last night and I almost feel like there's
something misleading about the example sentence we've been using; or maybe
I've been staring at it too long. Other sentences end up not even making
any sense in the FL2 languages except the one I design them for.

> I might not want to mention this in the article, but from a Finnish point
of view you could say that FL2 has two parts of speech: genitives and
locatives. This would be a matter of having a postpositional language where
the postpositions are actually adverbials:
>
> talon vieressä
> house-of beside-in
> in the beside of the house
> "beside the house"
>
> vs.
>
> vieren talossa
> beside-of house-in
> in the house of the beside
> *house the beside
>
> Note that verbs can also be inflected, although the root is basically a
noun: just one choice for a translation of "a see" is näkemä, and some
derivations in active use in Finnish are näkemässä (in), näkemästä (from
in), näkemään (to in or into), näkemällä (on), näkemältä (from on). I
remember reading a long time ago that proto-Finnish didn't even have
conjunctions although I doubted it then.
>
> So here's the old example sentence in "Finnish":
>
> [näkemän [talon [mäen päällä] akkusatiivissa] [pojan nominatiivissa]
puheenaiheessa]
> [seeing's [house's [hill's top-on] accusative-in] [boy's nominative-in]
topic-in]
>
> where all [-words are genitives or nouns in the genitive, and all ]-words
are locatives or nouns in locative cases (mainly the inessive).

If I'm understanding correctly, this would follow the 3rd language I
defined with those semantic grammars. (except of course that it's spatially
flipped). So FL2.4 isn't what you mean by FL2? (Let's call this option
FL2.3)

>
> I did this for the first time just now, but when you copy-paste the
"Finnish" sentence to http://ironcreek.net/phpsyntaxtree/ it gives you a
plausible FL1 tree where sub-sentences are read top-down, and the "verb" is
the root. It seems just perfect.
>
> Also important: don't forget that if you told a syntactician to analyse
the structure, without any introductory talk of the semantic formula, they
would do it like this:
>
> conjunction - subject - object with place adverbial embedded - verb
>

Maybe?

> The grammar makes no formal distinction between
conjunctions/prepositions, verbs/nouns and objects/adverbials, but that
doesn't strictly mean they aren't there.
>

I think if they're there at all, it's because you're using terms like
nominative and accusative which carry the grammatival theory with them. If
you restricted yourself to common English words this probably wouldn't
happen.

> > For example we can make the "is-a" one less flat by using the two
> relationship, "is" and "in", which sounds like it would match your
> original intent. Let's see if I can define it...
>
> > FL2.4 (I'm going to start naming these for easy comparison)
> S -> aXb: (X there is a, namely b)
> S -> SS:  S and S
> X -> S:     S, within which,
> X -> (empty): (empty)
>
> > Then we'd get:
> - ((There is nom, namely boy) and ((there is on, namely hill), within
> which, (there is acc, namely house)), within which, there is topic,
> namely see.)
>
> Sure, using equals (=) would seem to eliminate semantic ambiguity. But
are you sure that the 'on' is 'hill'? That would imply that the interior of
a house is a house.

Maybe you're not understanding my grammars. I am absolutely certain that
the "on" is "hill", because that's the definition of the language. FL2.4
can be summarized by saying [X Y] asserts X and Y exist and the Y is an X;
and additionally, if placed in a larger phrase, [X Y] asserts the larger
sentence occurs "within X".

>
> > I'll retroactively name this FL2.2
>
> That's what I'm talking about :D
> You've started making FL grammars! There are an infinite number of
possibilities, but I think some of them may turn out more useful than
others.
>
> > We can see that there is an acc which belongs to house, and within
> that acc, hill has an "on". This requires a different conceptual
> framework; a different definition of the words involved. The way I was
> thinking of it, "acc" is a role which the house has; it can play that
> role in different contexts of course, which is why we say the role
> belongs to the house; and it is only within the context of that role
> that "hill has an on". The "on" needs to function in a way similar to
> the "acc" and the "nom": since this language isn't claiming directly
> that the house is on the hill,
>
> True, it says that the house has a hill. I think this may be a case of
vagueness rather than ambiguity, and if you want to be more precise, you'll
have to mention in what way the house has the hill. In this case the house
would have the hill under it; it could also have a hill behind it etc.
Genitive doesn't have to be a matter of legal possession; you can for
instance have a home country without owning it.
>

I'm not sure what the difference between vagueness and ambiguity is. Is one
intentional?

The house doesn't have the hill in FL2.2. It has an acc which contains an
on which belongs to the hill. Which does have some useful ambiguity, yes;
for instance we can imagine the FL2.2 sentence

[memory [face girl] boy]

Here, the boy has a memory, and in that memory is a face, and that face
belongs to the girl (who presumably still has it). But we could also do

[memory [facial-expression girl] boy]

This could have the exact same semantics, but when I chose
"facial-expression" I was going for something temporary. So let me suppose
"facial-expression" is defined within FL2.2 in such a way as to constrain
"girl": because the facial expression is in the memory, the girl's making
it must be too; that is, she's not still making the same face.

Actually there is yet another possibility. I can't think of a great
example, but we can ask whether the boy remembers the girl or just the
expression. And we could leave that up to vocabulary, or put it in the
semantics of the language.

> >> I think both should be acceptable as they produce the same language,
right?
> >
>
> > I hope I've convinced you they do not! The intent of my post was to
> offer 3 different "FL2-type" languages.
>
> I was thinking maybe the formal language would look the same. I'm still
waiting for my copy of Languages and Machines…

These languages are about as formal as they come, except of course we're
not defining any of the vocabulary formally. As you said above, we could
replace the English "is" with a symbol; in fact we could use the set
membership symbol for "in" and maybe use functions to formalize "has"/"of".
I don't think it makes much difference though. Making them a little more
exact will only reveal more decisions to be made. The fact is, these are
different languages.

I mean, now that I've messed around with the formalized versions I can ask
a series of much less formal questions, some of which you've already
answered:

Does "[X Y Z]", acting as the prepositional phrase "in ...", modify the
sentence it's embedded in? Or instead, does it modify the sentences Y
embedded inside it? Or, does the language deliberately not specify, leaving
this to vocab?

In the sentence it's modifying, does it modify just the [-word, or just the
the ]-word, or act adverbially on the sentence, or some combination? Or, is
this left to vocabulary?

Does "[X Y Z]", acting as a sentence, assert that Z is of type X, or that Z
has X, or that Z and Y have X, or that Z and Y are a type of X?

I have a dozen or so such questions scattered in my notes. There are no
wrong answers, and different answers give very different languages.

>
> [Logan:] > To produce a useable language, you need to know not only what
order
> words can be allowed to appear in, and what parse trees those
> correspond to, but what *meaning* is associated with the syntactic
> structures produced by each production rule. That's part of what
> Daniel Demski has been exploring, and you'll note that he tried out
> several different possible formulations of the production rules for
> FL2 in order to try out different formulations of the semantic
> interpretation rules that need to be associated with them.
>
> Right, this is what I should be learning now.
>
> > If I've been following correctly, the syntax of FL1 is supposed to be
> given by S -> aS | <empty>; in other words "sentences are strings of
> words, one after another". Surprisingly, that actually does tell you
> More Than Nothing- it means that words can't occur interdigitated with
> each other (which is *not* the case in, e.g., iljena!)
> But if that's the best we can do for a minimal universal grammar, it's
> not very interesting. It's quite in line with what somebody else
> recently said in another of these threads on language universals- true
> mathematical universals are likely to be trivial and uninteresting.
>
> > If it turns out both that FL1 really is usable, and that the syntax
> really is that free, and it possible to give a sensible interpretation
> to every possible string of linearly ordered words in FL1, then I
> suspect we will find a great deal more complexity in the
> interpretation rules.
>
> I had the compounding semantics as a grammar rule last year, but this
time I omitted it - I'm now thinking of it as a convention rather than a
rule - although it can easily be added as a strict rule without making the
language too complex for any purpose. It seems to follow from the decision
of top-down vs. bottom-up reading.
>
> Other than that, you remember in the first message I said something like
"Here's a language that takes a minute to learn but a lifetime to explore"
although I was lying - there were two language (now called FL2 and FL1).
You can study their behaviour endlessly, but they're still based on the
given rules. It's as if a programmer codes an algorithm that generates
kaleidoscopic pictures endlessly; it's no use saying they couldn't have
made "all that" with "just these" command lines, and claim that they should
make the algorithm more complicated in the name of fairness!
>
> You see, the language is much simpler than natural languages because it
has fewer rules, but it's also much more dynamic for the very same reason.
For instance in English you can say "a cat caught a mouse in the barn", but
you can't say "a catch moused an in barn the cat". However the latter can
be a perfectly valid sentence in an FL-type language. This does not mean
you'll have to add all those rules to make it possible. It means you didn't
add all those rules in the first place, and that allows unusual ideas to be
grammatical.
>
> > One possible interpretation rule is provided by David Gil's IMA
> (Isolating-Monocategorial-Associational) framework- the association
> operator, which just says "the meaning some phrase followed by another
> word 'a' is something that is associated with both the meaning of the
> previous phrase and the meaning of the word 'a'." It is proposed that
> that is the ultimate basic semantic rule of all human languages-
> sticking words together makes new meanings that have "something" to do
> with the meanings of the parts- and everything else is just ways of
> narrowing down exactly what the "something" is.
>
> Thanks, I might be able to use that as a source. In FL1 you make compound
words where each word or morpheme modifies the one after it, and each word
is related to the one before it. In agglutination, words are put together -
there are famous examples of this in Welsh and German. In English they are
most often written as separate words; e.g. wildlife reserve + reserve
legislation + legislation session gives you the semantics of (wildlife
(reserve (legislation (session)))). That's basically FL1.
>

If you want FL2 to have any connection to FL1 it seems like you have to
have X act on Y and Z equally (again referring to [X Y Z]). This means you
should choose semantics that only have one relationship, either "in" or
"has". This could work with the relationship "is a type of" too, but it
would be very different from what you've described so far.

> > But all by itself,
> it's not really enough for a useful human language. The IMA model
> comes from an analysis of Riau Indonesian, but even Riau Indonesian
> isn't actually quite that simple- it does have extra bits of grammar,
> in the form of a closed class of function words, to narrow down the
> ambiguity.
>
> That's just perfect for my theory! Avoiding ambiguity is one key to
explaining _why_ things happen as they do. The way it would look good for
me is if there's one fundamental language that all languages are based on,
even toy grammars that accept any strings, and that's FL1. If a language
like Riau didn't have function words at all, it would branch off to a
different kind of FL2 language. But if any of the function words are case
markers, it's very likely to continue on the same branch with (all?) other
languages. I don't have enough knowledge to demonstrate the point where
languages like Riau and Salish will branch away from each other, but it's
doable with the FL method.

I wish you luck in working out the details of this! When you put it in the
abstract like that it doesn't sound impossible. I feel like FL1 and FL2 as
you've presented them are both too vague (ill-defined) and too specific
(semantically) for this. But what do I know.