Phenomenon-specific discussion of improvements to MRSs
Leader: Dan Participants: Petya, Rui, Yi, Petter, Lars, Ann, Angelina, Woodley, Francis, oe, Bec, Emily Scribe: Emily
Topic 1 (and only): Treatment of sentential subjects
To arrive bothers Kim.
Bother in the lexicon looks for a referential index for ARG1 --- typed to ref-ind to avoid expletive index. But the semantics of VPs is not a referential index. Tree doesn't give a rule that wraps referential index around the VP's event. Chose instead to do something on the other side -- pumping bothers Kim VP to a VP that is looking for a handle-valued ARG1.
Ann: Why do the pumping on the VP rather than on the subject?
Dan: Was there a compelling reason for that?
oe: There's a nominalization!
Dan: Yes --- the verb already says that the ARG1 is a referential index --- can't undo that.
oe: When a CP is a subject it becomes a nominalized argument, but as as complement it's not?
Ann: The structure we saw yesterday had a p in ARG1.
Dan: Yes, in terms of the shape of predication itself it's ready, but the valence list can't say it's underspecified between NP/VP[inf]/CP?
Emily: Why can't the valence list talk syntax only?
Dan: The selecting head has to know where to look in the HOOK to pick the right thing.
Emily: But you don't have that problem with complements?
Dan: Always bifurcate the lexical entries.
Ann: Why say p if you're not taking a handle as the ARG1 --- doing the nominalization --- why does the SEM-I say p? (I'd rather it didn't.)
Ann: The underspecfication of something as p (handle or ref-ind) always concerns me because logically they are very different things. I think it's probably a better semantics to think of it that way because there are some properties that are better captured by treating this as proposition.
Emily: Same with complements, or subjects are special?
Ann: Subjects are special.
oe: I believe your claim./I believe that Kim arrived.
Ann: There have been some arguments about this, but I am neutral. My feeling is that because the sentential subjects are relatively less common than the sentential complements and subjects are normally nominal guys, it's more reasonable to the conversion there. However, I can't unfortunately remember the literature with no notice … need to go back and look at it. I don't mind if you end up with a structure where all propositions are nominal could be interesting, but I believe this works.
Emily/oe: What about That Kim arrived was believed by every one.?
Dan: You're asking a tricky question. First you're asking whether the passive rule allows the construction of something that has the right shape but because the passive rule usually works on transitive verbs, then you're asking whether I can throw that NP expectation away and put in a clausal subject?
oe: [looking at demo] That Kim left was believed by everyone has the same semantics
as Everyone believed that Kim left. So it's not the subject, but the ARG1 that gets this special treatment, not the subject.
Dan: Right --- I had to add a separate lex rule for passive that explicitly allows for passives of sentential complement verbs. But it's right that we get an asymmetry.
Ann: But it's not at the semantic level because it's not ARG1.
Emily: But what's the motivation for treating ARG1 differently?
Ann: First, there are no verbs that only take sentential and not nominal subjects, but there are verbs that take only sentential complements and not nominals (collected by Grimshaw).
Ann: Secondly, don't want to go down the route of all thinks like if etc taking sentential complements doing nominalization. So we're going to at least have a class of things doing what we do currently (handle-valued ARGs) but arguably also a class that do nominalization. And the main reason I think I want these is to avoid the p-type underspecification.
oe: So exploiting the p in the ARG1 position makes you especially nervous?
Ann: Because it's all over the place. Because every time we see an MRS with a verb in we'd have to do something to that MRS.
oe: I see the difference in frequency, but is there a difference in principle? Is the difference in frequency enough to motivate absorbing the lack of parallelism.
Ann: I have some recollection that sentential subjects are different from sentential complements in at least some sense.
Woodley: Was talking to Ann on Monday about scopal arguments and how to deal with them --- reifying propositions.
Ann: That's one account of how to do the interpretations of MRS.
Woodley: Are there cases where the sentential subject position could be interpreted more like the modals where you don't want to do the reification?
Emily: That Kim left is possible./Parties are possible.
Dan: I have two adjectives possible, one which can take a sentential subject --- there's a class of predicates that take expletive it or sentential subjects.
Emily/oe: So what about bother and surprise?
Dan: No nominalization there. That's because the base entry for surprise that take an expletive subject and a sentential complement.
Ann: Is there another surprise for Kim surprised me. or is it going to be the same rel?
oe: Or should it be a different rel?
Dan: from SEM-I:
"_surprise_v_1_rel": ARG0 e, ARG1 i, ARG2 u.
Ann: Should be two separate predicates.
oe: Is that different senses?
Ann: We have other cases where we have two predicates differentiated only by the set of ARGs. And I don't see that we're helping anyone by having predicates
Woodley: That Kim arrived surprised me. seems different from That Kim arrived is possible. The latter seems like something that can happily be treated as a logical operator, but not so much for surprise.
Dan: So what do you want for It surprised me that Kim left.?
oe: Just to be difficult, I think that It surprised me that Kim left and That Kim left surprised me are perfect paraphrases (and should be represented as such).
Ann: Davidsonian … quotes … [missed this one] … there does seem to be a difference between that and surprise.
Dan: There seem to be pretty close paraphrases in That Kim left surprised me. and The fact that Kim left surprised me. But It surprised me the fact that Kim left is different (if grammatical).
oe: Can't you do the the fact that paraphrasing with most sentential objects?
Dan: Only if the verb can take an NP as well as a CP.
Ann: Presupposition difference (because believe isn't factive, but surprise is).
oe: So could paraphrase with the proposition that or the claim that instead.
Ann: Are there cases where we have non-factives for sentential subjects (without passive)?
Dan: whether complementizer: Whether or not Kim left did not emerge from the courtroom evidence
Ann/Dan: Complementizers might be semantically significant (per Davidson) but might be just a syntactic idiosyncrasy.
Dan: We treat that as a syntactic fact.
Ann: What semantics do you give for Whether Kim arrived on time emerged.?
Dan: wh-subjects always turn into NPs, so the nominalization is there.
Woodley: Does that predict that it doesn't go with predicates like possible? *Whether Kim arrives is possible?
Dan: Wh-clauses have the distribution of NPs, where that-clauses don't. So always pump wh clauses to NP, and don't want to pump all that clauses to NPs, since then they have to be constrained away from lots of other NP complements.
Ann: But if semantically you don't think that the wh things are that different from the that things, then that would be an argument to treat them as nominalizations in subject position.
Dan: This brings us back full circle to the thing I'm not too proud of: I want the extraposition symmetry preserved, I wanna do something sensible with sentential subjects, and … [EMB: maybe this third one was wanting consistent treatment of clausal arguments in general?]
Guy: Does the extraposition/intrapositon rule also apply to VP subjects? to arrive bothers me
Dan: It looks like no. Do I get It bothered Kim to have to buy the book.: ERG currently doesn't get the analysis where VP[to] is a complement of bother. I probably should have another lexical entry for bother in this frame (with expletive subject). [Searches for valence frame --- no verbs like that yet.]
oe: In that case it would pattern with surprise.
Emily: and that-CP taking bother, in fact.
Dan: Can you think of four other verbs?
Guy: Would that give an additional analysis for to add lexical types bothers you?
Dan: If I've written the existing lexical rule type correctly, it'll come for free. The power of linguistic prediction!
oe: But this is increasing the number of cases of that discrepancy between reified and non-reified ARG1.
Dan: For some strange version of counting … it preserves the generalization (about expletive subject alterations).
oe: But we got interested in surprise because of that asymmetry.
Ann: For the expletive cases, I think this is right.
Ann: The symmetry for the extraposition alternation is more important. In all other cases you're talking about adding lexical entries, which gives a place for putting in two different predicates.
Woodley: Direction of lexical rule: if it's non-extraposed to extraposed, that would potentially make it possible to be consistently nominalized.
Dan: Ann is saying that the semantics of the extraposition cases is not up to debate. *It brings up several problems that Kim left.
Woodley: I star your star.
Emily: But are we saying that the two frames have to be consistent, or that they may not have nominazliation?
Woodley: Can you knock down the straw man that any sentential subject can be extra posed?
Dan: That Kim sings songs vanished from our awareness./It vanished from our awareness that Kim sings songs.
Ann: Let's stick with what's in the corpus…
Dan: Nothing rides on this except the empirical facts. Depending on that distribution that might change the lexical rule that takes any verb and
Dan: It persuaded John to leave the store that Mary went away.
Emily: If you did need that ugly rule, does it change what you want to do about nominalization?
Dan: No. Or maybe it does. Then we could always pretend that a sentential subject is the corollary of the
oe: Suggestion for homework from semantic documentation point of view to possibly both of you (= Dan + Ann?). At this point it seems like there is the possibly that it is more parsimonious to use the underspecification mechanism also on ARG1s and say the choice of reification v. not happens later. You (Ann) said that there may also be literature on the difference between sentential subjects and complements, which might also inform us. So maybe we should turn to that next.
Ann: You would think that whether is okay with a nominalization or not?
oe: I'd currently like to entertain the possibility of no nominalizaitons?
Dan: Then I'd have to have prepositions also into taking hole arguments --- wh guys are NPs and get nominalized everyone. No like. That's not an argument, it's an emotional indication of …
Glenn: … the work getting done.
Dan: More research is needed. It would be fine if people brought illumination to the discussion.
Francis: That the results were inconclusive surprised no one.
Emily: Shall I record that as use or mention?