Print

Print


On 10 September 2015 at 20:05, And Rosta <[log in to unmask]> wrote:
> Some very brief feedback, in the uncertain hope that it is better than none:
>
> 1. What leads you to devise devices, such as nonintersective adjectives,
> whereby the syntax is not homomorphic with the logical form? Various answers
> are possible, e.g. "whimsy" or "ergonomic considerations (e.g. brevity,
> processability)", but at any rate I find myself feeling a nead for an answer
> to that question before I can assess the device itself.

Well, that's complicated. I would argue that a syntax for
non-intersective adjectives that just involves replacing referents
from left-to-right (or vice-versa) *is* reasonably homomorphic with
logical form, insofar as the logical form is also just a linear,
non-branching graph.

But that argument aside, why want non-intersective adjectives at all?
Well, for one "because natlangs have them", and being able to
translate them into the unnamed FL1-inspired language significantly
increases its unambiguous expressive power (since it can now encode
centralised webs with limbs of arbitrary length, instead of being
limited to 1 step from the hub). WSL already has methods of expressing
arbitrary trees via embedded clauses, so for the question of "why want
them in WSL", the answer is indeed "ergonomics". It's an extremely
common special-case structure, so it makes sense to have briefer
special-case syntax to handle it. The need for whole embedded clauses
to express multi-word genitives in WSL has been bugging me for a
while.

> 2. Nonintersective adjectives are semantically heterogeneous; all they have
> in common is the lack of intersectivity. I suspect your device wouldn't
> generalize to all, but I can't check that, since I don't understand the
> device enough and I don't have handy a taxonomy of nonintersectives.

They are indeed semantically heterogeneous. And it's possible that
there may be some other semantic features that are also hard to
represent for other reasons. But if there is some non-intersective
adjective that is still difficult to translate because of other
semantic features than simple non-intersectivity, that is not a
failure of the device meant to handle non-intersectivity. The same
argument could apply to regular intersective adjectives and nouns-
they are semantically heterogeneous, and there may be some that are
still difficult to represent despite not being non-intersective.

If I encounter any, then it'll be time to come up with new devices to
handle whatever the new problem features are.

> 3. To many of your posts I can't give the attention they deserve, because I
> don't understand enough on first reading and, especially now that the new
> academic year is starting, I don't have the time to devote the extra time to
> puzzle out more of them. I say this by way not of reproach or complaint but
> rather of apology for not doing your posts justice.

You have expressed this before; 's OK; it's nice to know my pondering
are of interest, even if they don't get a whole lot of response.
However....

> 4. First-reading grockability would be enhanced by many more examples.

This is quite true. Examples take time, and so I am disinclined to
write out a bunch of them when just trying to get the basic idea
*written down*, but I'll try to give it more attention in the future.

In the meantime, for something a little more concrete, this is the
relevant modification to the interpretation rules for WSL:

[|N: n N|] = λy.[|n|](y)([|N|]) i.e., the denotation of WSL Nouns now
takes two arguments- the variable bound to the referent that its
describing, *and* the denotation of the remainder of the noun phrase
(the degenerate case, where there aren't any more following words, is
[|N: n|] = λy.[|n|](y)(λy. true); the constant value "true" can be
simplified away later in the derivation).

Normal intersective Nouns, like "blue", would then have a structure
like λy.λz. blue(y) & z(y); i.e., it's just the logical conjunction of
the basic predicate behind this Noun and whatever the denotation of
the remaining phrase is, operating on the same variable. Just like
before, except that the conjunction is lexically specified, rather
than being part of the meaning specified by syntax. (The ordering of
Nouns within a phrase is significant now, though, so syntax is still
carrying pretty much just as much information as before- just
different information.)

Non-intersective Nouns, on the other hand, have a lexical semantic
structure like λy.λz. ∃w. former(y, w) & z(w); i.e., there is some new
entity which the rest of the phrase will describe, and the one we've
been talking about up till this point bears some relation (in this
case, "former") to it.

What just barely occurred to me while writing out those examples is
that it would be possible to lexically specify disjunction rather than
conjunction, or any other logical relationship, really. I probably
won't make use of that in WSL, but that suggests that different kinds
of conjunctions can be worked into the monocategorial language still
without having to actually introduce any new syntax or lexical
syntactic classes.

The operation of non-intersective words in the monocategorial language
is essentially the same, but the semantic formulas are a bit messier
since they've *already* got to keep track of 3 variables at a time.

-l.