(Discussion on types of modifiers)

Emily: The pointer thing is really relevant. HOOK contains pointers to things in the RELS list. It maintains access to information. I don't think it's wrong to think of directionality, so long as you're not getting confused about unification.

Zhen Zhen: And scopal ones have ARG1 pointing to LTOP?

Emily: Arg0 of the modifier is not identified with anything else. The INDEX of the modifier is the same as the index of the head. Scopal modifiers are not having their own index point to the modifier because we want the HOOK information to be passed up the tree along the semantic head path.

(discussion of a syntax trees in Chinese: "very not good" "not too good" and what types of modifiers they use)

Emily: If we were actually doing scopal modification of nouns, which the ERG doesn't bother with, you would get things like "the likely winner lost", where you need the index of winner to be the object(? subject?) of the verb

In (Chinese sentence with) "not too good", we want "too good" to unify first, with "too" as the specifier for "good".

And in "too not good" Emily suggests the middle one is a degree modifier.

Zhen Zhen: Sometimes ARG0 and INDEX are identical, and sometimes not. ARG1 is sometimes identified with itself.

Emily (finds a type in the ERG): It links its specifier's XARG and its ARG0. The specifiers say "My XARG is my ARG0." This is not so much about passing out, it's the degree specifiers giving a pointer into their predication, and the adverb taking the degree specification links its own ARG0 to the degree specifier.

What's unusual about this is that we have a head daughter that is not a functor. "My XARG is my ARG0 (ARG1?)" is all over the place. It's the pointing outside that is unusual.

You should be able to do the same thing with your scopal adverbs so they can hook up with their degree specifiers.

If you have cases in Chinese where there's clear examples of intersective modifiers attaching outside of scopal modifiers, I am very interested in that.

I'm also interested in how you get "a not too good book" to have the same label of "book" as in "good".

The intersective modification rule says the LTOP of both daughters is the same. Having a rule saying the LTOP is the same in this case is probably wrong, you probably want it lexically on the daughter types.

Zhen Zhen: LTOP of "book" is the same as LTOP of "good," and not the LTOP of "not"?

Emily: I think so. That's what I said. Now I'm not so sure. (investigation occurs). I take it back. Maybe we're not passing up the LTOP on purpose. Having talked through all that, I'd say we have an argument of having the rule say the LTOP is the same.

We have a QEQ between "not" and "good," which in theory means we can shove a quantifier in there. (Something here about how quantifiers can happen inside or outside)

Francis: So that's where we need to look?

Emily: Yes

Francis introduces a new question about "This classifier that I read book" vs. "That I read this classifier book." (It sounds like the indices are not ending up as they should.)

Emily: The way we do NPs that the determiner effectively seals of the scopal information in the N-bar. I wrote a paper on this on Turkish: "Semantic Construction for Nominal Expressions in Cross-Linguistic Perspective." (Bender, Egg, Tepper 2006)

So this is a known problem, and it boils down to the LTOP of "book" no longer being available to intersective modifiers outside the scopal modifier. This is part of a more general pattern of problems.

Francis: One of the problems is that we get a parse that looks right except for those indices co-indexing. Do we just live and let live?

Emily: Yes, I'd say so, as long as you document it.

Zhen Zhen: Is there a paper on this?

Emily: There's not a right answer out there that's known. There's what's been developed in the various grammars and we're documenting it for the ERG.

The hard part in grammar engineering is deciding what you want your representations to be and then getting there. Don't make the mistake of thinking that there's a right answer for everything out there.

The people doing formal semantics have never looked at it in the sense of a broad coverage grammar.

Zhen Zhen: (Question about (parallel) conjunction structures in sentences like "C1_______, C2_______" "C1______, adv________")

Emily: See what Dan does with "if, then" constructions. There's a rule that takes two daughters, and you're probably going to use the MINORS feature. Each of the paired conjunctions say what their MINORS value is so you can get the matching pair. Then you have the problem of how you get that information up to the mother. The conjunction case is easier. The adverb case is harder, but not impossible -- it can say "my MOD's value is (what we need)."

In "adv,    ,     ", (a sentence like this but with an adverb) the adverb can pass the MINORS value up to the head daughter. It's not impossibly deep. The rule at the top says the MINORS value must be the same.

Two things in MINORS: the type of the thing, and whether it's first or second.

LadUW20161110 (last edited 2016-11-10 20:07:31 by LaurieDermer)

(The DELPH-IN infrastructure is hosted at the University of Oslo)