Discussion: Scopal and Non-Scopal Modifiers
Kim didn't speak for a long time. Kim deliberately didn't speak.
Dan: didn't speak for a long time ex is quite a nice case study. Does it make sense to say yes it's ambiguous and the for along time takes as label the neg_rel but takes as ARG1 the ARG0 of event of speaking. If we want the ARG0 of the neg_rel, that forces it out into the open, which we haven't wanted.
Guy: Ann says it seems strange to talk about a negated event, to have negation itself is a SOA.
Dan: not speaking is fun --- what is the thing that's fun? What's the argument of fun. Not just speaking, it's the non-speaking that's fun.
Emily: But if not speaking is a nominalization, then we have an 'x'.
Guy: But that variable is just as dodgy as the negated variable. We have to justify one of these variables.
Dan: If you move out of the MRS into the real universe, what are you saying isn't fun?
Guy: In the model of the world.
Dan: There needs to be some state of affairs we're talking about.
Guy: What is a state of affairs that is a nominalization.
Dan: Maybe this takes us too far afield.
Guy: But maybe it's an argument if we have to be able to refer to it.
Emily: Can it be the antecedent of anaphora? Someone else think of it while I'm typing.
Dan: John didn't speak. It really surprised us because John always speaks.
Emily: But we know that it can refer back to propositions.
Dan: *John didn't speak and Kim did so.
Woodley: *John probably spoke and so did Mary.
Oe: John frequenty visits and so does Mary. --- Is frequently scopal?
Dan: It satisfies the entailment test.
Berthold: A presupposition attached to so... do we represent negation as events.
Dan: If it's really there, then you ought to be able to pick it out with an anaphor like so.
Woodley: Can you pick out the event for blue in A blue dog just walked by with an anaphor? Maybe it's too stringent of a test.
Dan/Glenn: John didn't speak, neither did Mary.
Emily: So what's going on there? ... refers back to event inside the negation.
Berthold: Making neg transparent to INDEX of complement --- neg LBL shared with intersective modifier (in high attachment), with ARG1 = ARG0 of verb.
Emily: That's what happens now, if we allow high attachment.
Woodley: But that doesn't work so well for Kim deliberately didn't speak. Are we really happy with ARG0 of speak as ARG1 of deliberately?
Emily: What I'm missing is a connection between our representations and the actual interpretations/things we can test like entailment.
Guy: Depends on what you want to do with negation in the object language. If it's just logical negation, it has to apply to propositions, and that doesn't help here.
Dan: Kim deliberately failed to speak --- not just a fiddling with scope. failed introduces a new predication with speak down inside. There fail supplies us with what we all assume is a predication. Somehow the action that was constrained was the speaking one, but the deliberateness is about the failing.
Guy/Dan: If we can talk about failing events, maybe we can talk about negation events?
Francis: Kim failed to speak and so did Mary. --- that one's fine.
Glenn: The do support?
Woodley: Something just weird about English.
Emily: Which do support?
Glenn: The one in the first clause.
Emily: But that one's semantically empty.
Woodley: Maybe it's not?
Dan: *Kim wasn't happy and Mary was too. It's not about the do support.
Dan: Kim failed to be happy and so did Mary.
oe: Kim never danced. So did Mary. Star?
Guy: Failing seems to imply more active involvement.
oe: Kim almost fell. So did Mary. Also a scopal modifier.
oe: Kim probably will turn out famous. And so will Mary. Maybe we're putting too much work on the so anaphor.
Dan: Then give us another test.
Woodley: Kim deliberately didn't speak. / Kim managed not to speak.
John: Deliberately not a great choice of word because with the post-posed variant, there's also the deliberately as slowly version.
Emily: [Tries to explain the current ERS.]
Woodley: Can you give me a world scenario where one is true and not the other.
Guy: In terms of intention to speak and whether or not speaking happened.
oe: Emily just misused the entailment test. It only goes one direction --- entailment doesn't show anything. So we don't know that deliberately is non-scopal.
Emily: Can we construct a slightly more complicated example that gives us different scope possibilities.
Guy: Scope possibilities is a separate question from denotation.
Woodley: I'd like to be able to find differences that aren't just lexical semantics but have to do with the structure.
Emily: My point was that the denotations might be world-equivalent, but allow different flexibilities in more complicated examples.
oe: Kim successfully didn't speak. Kim succeeded in not speaking.
Dan: But there's a nominalization there.
oe: Play the same trick with a verb that actually selects a VP and has a closely-related looking adverb.
Dan/Guy: Back to big questions:
- Is the denotational distinction clear in practice?
- Is the scoping machinery adequate?
Woodley: Do we have a collective sense of whether or not deliberately is scopal?
Emily: And how would we know?
oe: Clear scope ambiguities might help...
Guy: That would assume that the scoping machinery is correct in all cases.
oe: You're pulling many rugs out from under us.
Jan-Tore: If heavily is intersective, shouldn't it's raining heavily also entail it's heavily. What about subsective modifiers?
Emily/Dan: We're trying to move to scopal/non-scopal, but we have intersective as a label in the grammar (incorrectly).
Guy: In terms of what the MRS gives us, it's saying that once you've expanded the scope tree there's some event that is raining, and some event heavy that takes it as its argument. Once you've quantified the event variables, then you have a complex predicate which you could be intersecting with the rain predicate. We're intersecting the things that are raining events and the set of things that have an associated heavy event.
Glenn: Would it be better to say the set of events that are heavy?
Guy: That's not what we have in the MRS. We have two events --- raining and heavy. We have the second variable in case we need to modify it again. This is not a move that every semanticist makes.
Emily: What about Kim probably baked a cake for Sandy. / Kim baked a cake probably for Sandy. --- seems like a surprising generalization of degree specifiers for me. These examples came up because I was looking for cases where the label of bake and for might need to be different.
Dan: So what do you want the representation to be?
Guy/Woodley: probably as a scopal modifier of for, probably shares label with bake, qeq for label as ARG1; for takes ARG0 of bake.
Dan: And Kim baked a cake not for Sandy?
Berthold: The most probable answer to our problem is X./The answer to our problem is most probably X. Different in the ERG.
Dan: The ERG implements the hypothesis that there are no scopal adjectives.
Guy: Is that right?
Dan: Show me some tests. Alex Lascarides says there aren't any. Treat them as the same. If that's wrong, then show me the tests.
Jan Tore: Entailment.
Guy: Is a probable solution a solution?
Jan Tore: Certain non-intersective adjectives will fail, but they're not scopal either. Earlier we said that that test is just one way, but adjectives are a counterexample.
Emily: I have in the past convinced myself that there are enough readings of The most probable winner of every medal was disqualified that we need the every to be able to go in between probable and winner. But this is the kind of test you told me I couldn't use.
Woodley: Question is if stone lion commits you to its being a lion.
Dan: Did you see the stone lion in front of the bank? That's not a lion, that's a tiger
Francis: Ham sandwich.
Dan: It's not a ham sandwich it's a lion.
Francis: Coercion from lion and representation of a lion.
Dan: You want it to be ambiguous?
Guy: Diff between metonymy and literal meaning.
Francis: A stone lion is not literally a lion.
Woodley: former is better. Saying a former X isn't committing to it being an X.
Glenn: Correcting intensional contexts is kind of out of scope here.
Guy: So the answer to the first question is no --- the denotational distinction is not clear cut. Not a very clear test.
oe: Are there really people who think stone lions aren't lions?
Francis: Me! A representation of a lion is not a lion.
Dan: What about a stuffed lion?
Francis: Depends on whether it originally was a lion.
Emily: Then it's a former lion!
Emily: But even if we find clarity in the literature on this, we still have the problem that the test goes in only one direction.
oe: Monday's solution to the problem turned out to be wrong. It wasn't a solution actually, but it was one on Monday. By that same reasoning, a former politician is in the set of politications, but the set membership has spatio-temporal extent?
Woodley: So does the event of not speaking have the predicate speak be true of it?
Guy: How could that be true of a non-speaking event? If we say someone didn't speak, we're not committing to them speaking at some other point. (Unlike with former).
Woodley: Exactly. In order for these to be parallel, you have that strange situation ... unless you're talking about strict logical negation, you're going to have speak(e) when you're talking about not speaking.
Guy: We're not committing to there being a speaking event.
Emily: Speaker commitments not every bit of the MRS, but calculated over the whole thing.
Guy: Not doing quantifiers for event variables yet, because don't yet know how. But if we treat them the same as every other variable, we'll need to quantify them somehow. So just because there's an event variable in the MRS, doesn't mean we're committing to it.
Woodley: Right. Kim didn't speak for a long time with high attachment of for a long time taking speak's ARG0 as its ARG1 but sharing LBL with neg, then the quantifier for speak's ARG0 has to be outside neg as well.
Guy: Create new predicate not-speak that maps from events to truth values.
Woodley: If not-speak is an opaque predicate...
Guy: It's compositional. We have to have some way of mapping from one kind of truth conditional function to a new one. Could be as simple as flip the truth value. When ever speaking is true, not speaking is false, and vice versa. Does require committing to being able to talk about not speaking events, and in this case it's also a for-a-long-time event.
oe: In this case, there's something in conventional logic that looks very tempting. But what about almost in almost succeeded --- much easier to start talking about states of affairs that are the almost ones.
Emily: So you're saying that we're being misled by the similarity between logical negation and natural language negation.
Emily: That brings up another question that I have: Why do we have a difference between heads that take scopal arguments and promote their own ARG0 as index and those that attach as modifiers which are transparent.
oe: Are shy about their intrinsic variables.
Guy: That's about composition, which I was trying to avoid here.
Emily: But also about representations, because some of the possibilities we are considering here are only made possible because of that transparency.
Francis: Back to stone lion. Nick Asher shows interesting facts from distributional semantics.
Berthold: But compounds might not be syntax...
Guy: Not clear-cut.
Woodley: So what's the alternative?
Guy: Yeahs, seems like a shame to get rid of info that we have.
Glenn: If there's a vilage of lions and there's one who sells stone, he'd be the stone lion ... and he actually is a lion.
Guy: On the second question --- we agree that we want two readings for Kim didn't speak for a long time but not yet sure whether we want to have negation events, or whether natural language negation should be the same as logical negation. Maybe almost gives better examples.
Woodley: Kim almost spoke for a long time / *How long did she almost?
Dan: Are we getting to a point where we see a very deep divide between Kim never spoke and Kim rarely spoke. The second one entails that Kim spoke at least once. But they seem like they're very close. Do we suddenly flip to a different type of modifier. Is rarely a different kind of modifier from not.
oe: These could all be scopal.
Dan: But what about frequently --- where in the cline are you going to make that cut.
Glenn: I don't think we can avoid treating zero as a categorically different thing.
oe: rarely and frequently come out on the other side of the entailment test than not and never, but they could all still be scopal.
Guy: Here's the slippery slope. Make everything scopal?
Dan: almost never means about the same thing as rarely.
Francis: It's a slippery slope. We know it. We don't know how and where to draw the line, but would like some reason to draw it.
oe: Or not draw it at all. Guy is dangling that possibility in front of us.
Emily: So trying that out and seeing where we get tangled up might lead us to a test that goes the other direction.
Guy: Comes back to Kim didn't speak for a long time
oe: Doesn't Emily's thought experiment fix that for us? Two readings.
Dan: What about Kim didn't speak loudly.
oe: Two readings.
Dan: What's the other one?
Francis: [Doesn't speak, loudly]
Emily: The silence was deafening.
Dan: I'm sure that's what that sentence means.
oe: But we're in the habit of giving more scoped readings than are easily teased out in the world.
Emily: Just a few...
Continued as SIG on Friday:
oe: Examples like I'm happy to discuss/We happily discuss: These are very close, do we want them to be so different in the ERS?
Emily/oe: Both attribute to each other the thought experiment of treating all adverbial modfiers uniformly, as scopal.
oe: If someone is trying to give us an MRS to generate from, this can be hard to predict if we can't describe the distinction.
Emily: Isn't this info in the SEM-I?
oe: Currently we have modifiers that have both scopal and non-scopal entries.
oe: Haven't paid enough attention to Guy's third item from last time. There's a syntactic perspective/folk wisdom that only one of the two types extracts.
Happily, I understand the dog is barking. -- ambiguous.
Guy: Quickly, I understand the dog is running -- odd.
Emily: Yesterday, I understand the dog went running. ok
oe: But we know that the time ones are scopal: Kim didn't speak for a long time
Emily: Scopal vs. non-scopal is different from high vs. low.
Guy: High attachment of non-scopal is not algebra-compliant.
Woodley: Possible coincidence about the adverbs that we've been looking at. The ones that we seem to want to be non-scopal typically have the property that they can be interpreted as referring to the subject.
Emily: In what sense does yesterday refer to the subject?
Woodley: Not talking about those yet. Take happily. Happily, I will eat chocolate.
David: But Happily, my dog wasn't sick.
Emily: Here's an example where we probably need two entries. Happily the speaker stance adverb (probably scopal, on current distinction) and the one that you say refers to the subject.
Woodley: So maybe there are three.
Emily: Three?? That was one of two for me.
Woodley: Happily, the dog didn't get sick., The dog was happy to not get sick.. The dog happily did not get sick.
Emily: Only two of those are happily
Woodley: But maybe they're the same predicate.
Guy: Try a different word with less ambiguity: fortunately. Fortunately, it didn't happen. It was fortunate that...
Woodley: Quick is unambiguous the other way. I was quick to eat chocolate.
Emily: But for happily you're trying to merge a non-scopal use with a raising or maybe control predicate, which must take its argument as qeq.
Woodley: I'm saying that in I'm happy to eat chocolate, it's happy(x) /\ and eat(x,chocolate).
Emily/David: That's not what that meants!
Guy: I'm happy to eat chocolate, but I didn't get to.
Woodley: But we have other irrealis things: I would eat chocolate.
Guy: But there's a modal there.
Emily: Also not a test, but it would be unusual to find a VP complement that is not a semantic argument of the (semantically contentful) thing it's a complement to.
Guy: What was the original intuition?
Woodley: The non-scopal ones seem to be the ones that are predicated-ish of the subject, so maybe that's a test?
Emily: The canonical non-scopal ones are manner, place, time, and then maybe secondary predicates.
Woodley: Not sure about time, but can't manner and place be about the subject.
[ Work on finding a plausible example to pull apart the cases where the subject and the event are not necessarily in the same place. ]
Emily hit Stephan with a ball over there. Object-oriented programming was invented over there.
David: And with expletive subjects:
It rained yesterday over there. It annoyed Kim in Oslo that the trains were late. It was beaten to death at DELPH-IN that...
Emily: So I think it was a red herring to try to have the non-scopal modifiers attach to the subjects.
oe: Guy was effective in the earlier session at getting us to focus on models, and not get tangled in the syntax/semantics interface.
oe: But back to syntax, you had another test?
Guy: The position of the adverb in the clause. The dog barked, probably (pause required). Dan currently doesn't allow that in the ERG with or without the comma.
oe: So The dog barked never shouldn't parse. And this is the case.
Emily: Is that bad? Maybe archaic or poetic.
oe: Even higher agreement for *Never the dog barked.
Emily: There's some special things going on with negation. not is very special, negative inversion: Never did the dog bark.
Guy/David: But: Quickly did the dog bark.
Emily: So maybe the negative ones require inversion, while the others allow it optionally (if archaically)
oe: Create table for testing positions, using set of adverbs ranging from pretty certainly scopal to pretty certainly non-scopal.
___ Kim knows that ___ the dog ___ will ___ bark ___ 0 1 2 3 4 0 1 2 3 4 adverb - - - - - not - - + + ? never - + + + ? probably - - + + ? almost - + + + ? allegedly ? + + + + rarely often ? + - - + for a long time + + - - + in Paris + + - - + here, + + + + + now ? ? ? + + quickly ? ? ? + + loudly NB: 0 is + only if the adverb can be intepreted as modifying bark.
Woodley: How about In Paris, Berthold proposed that we hold the 14th summit.
David/oe: False! It was in Olso!
Emily: This is the kind of construction that is very discourse-sensitive, so easy to star something that then turns up the corpus.
David: Contrast can help.
Woodley: Emily proposed that ACL be in Paris. Berthold wanted ACL to be in Madrid. In Paris, Berthold suggested we have DELPH-IN.
All: Yeah, that's fine.
Guy: Can we play the same game for the scopal ones?
Woodley: Rarely, Woodley claimed that would adverbs not modify their subject. Other people disagreed. Others suggested that this ws in fact common. Rarely, they said/maintained adverbs would modify their subjects. ... hard to tell with the quotative in there.
All: Less convincing.
Guy: Emily thought that the dog probably yowled. But Berthold thought the dog certainly yowled and probably Berthold thought the cat yowled. ... only getting high attachment, so here's a contrast.
oe: Emily wanted to discuss scope for a long time. Berthold wanted to discuss scope briefly. For a long time, Berhold wanted to discuss Hausa.
oe: Kim likes to foxtrot slowly. Sandy likes to foxtrot quickly. Slowly, Sandy likes to waltz. That's pretty good!
Woodley: You want it to be good.
Guy: It's understandable...
Guy: here, now, in Paris are just placing the event, where quickly, loudly, for a long time are adding to the description of the event.
David: Kim knows that probably here the dog will bark. Must be [ probably here ]
Emily: So if there's only one slot there in position 1, that gives us an argument for it to be extraction.
Guy: Kim knows that now here the dog will bark.
oe: That's spatio-temporal clumping, a known phenomenon: On the bed in the afternoon is a good hiding place.
oe: So at least we've gathered some ... facts. I expect that H&P will talk about scopal and non-scopal.
Emily: I don't. I expect them to have 6-7 categories, but it's worth looking at those cateogeries and trying to align them with our scopal/non-scopal.
Woodley: Most robust distinction seems to be position 4 being available for putative non-scopal only.
David/Woodley: *Kim knows that the dog will bark never. Kim knows that the dog will bark almost never. ??Kim knows that here rarely the dog will bark. Kim knows that here rarely will the dog bark.
Emily: Is rarely temporal enough to participate in the spatio-temporal clumping?
Guy: Example where EDS has lost too much info reconstruct:
I met the person whose happy dog is healthy vs. I met the person whose healthy dog is happy
oe: The only remaining difference is that one is tensed and the other one isn't.
Woodley: So do it with verbs:
I met the person whose dog that barked smiled. vs. I met the person whose dog that smiled barked.
[ This didn't parse in 1214, but does parse in trunk. ]