When pied piping is optional: Some ambiguity vs. lack of some coverage?

While doing some final testing on fronting languages, I noticed some problems in my treatment of optional pied piping. Specifically I was not covering determiner extraction from subjects:

(1) какой  приехал человек?
     which arrived person?
`Which person arrived?' [rus]

This was because the basic-head-subject phrase was not allowing slashed non-head daughters. Relaxing this constraint requires instead constraining subject-head, so that there isn’t ambiguity in sentences like (2):

(2) какой  человек приехал?
    which  person  arrived?
`Which person arrived?' [rus]

Similarly it is tempting to constrain complement-head (to insist on slash-empty non-head-daughters), to avoid ambiguity in sentences like (3):

Какой сон   видит Иван?
which dream sees  Ivan
`What dream is Ivan seeing?' [rus]

However, I then lose (4):

(4) Сколько  Иван книг  читает?
    how.many Ivan books reads
`How many books is Ivan reading?' [rus]

So it looks like generally, the solution is not the right one.

It sounds like this may be similar to the problems we were solving for the head-adjunct and the in-situ rules.

I wonder if I can apply a similar solution here? Any thoughts?

Is there a reason that this kind of ambiguity is so troubling? To me, it just seems like a potential efficiency issue (and a relatively minor one).

Perhaps the ambiguity would be desirable, if there is a subtle information-structural difference, perhaps correlating with some intonation difference.

The lack of coverage seems like a much more serious problem.

This is what I thought too, but @ebender is pointing out that it is the information structure library which should provide this coverage, not me. BUT, the information structure library does not in fact support free word order! So… Perhaps that’s it, that’s the support?

Right, and keeping the ambiguity will make it easier for someone in future to extend the information structure library to cover free word order.

1 Like

I am wondering if, ultimately, fronting generally should be covered by the info-structure library, while the question semantics is additional and is not part of the fronting rule.

It never looked to me like the additional filler-gap rules which I added for the constituent questions library will play nicely with the additional filler-gap rules which the information structure library adds. But I couldn’t quite figure out the relationship between them. The infostructure library uses L-PERIPH feature heavily and I think that it may be precluding sentence embedding that way (not 100% sure at this point). But other than that, it always seemed to me that there may be duplicate work that I am doing here, with the filler-gap rules to cover question-specific fronting.

Or maybe not, because focus may be sentence-final in fronting languages. I remember @sanghoun saying that it may make sense to say that questions are contrastively focused, but I that doesn’t quite match my understanding of what contrastive focus is? Or perhaps we just need a more general definition: a type of focused position which is used for e.g. contrastive and question-related focus. Then something like Russian can be modeled:

  1. Focus is sentence-final, but we don’t know what to do with free word orders, so, we are not modeling that anyway.
  2. Contrastive-focus is sentence-initial, and we add a particle that marks it. Again, because we do not know yet what to do with free word order, the information structure library won’t add much at this point.
  3. Question-related focus is like contrastive focus (and also uses a particle! a different one but still).

Item 3 is covered currently by my analyses for questions, and I suspect that the analysis I added might also cover item 2. I could rename the wh-ques rule to not mention wh, and I could move the rule in the info-structure library, and leave everything else as is. Then item 1 remains unimplemented for free wo for now, but we will have 3 and possibly also 2. How is that for a plan? It will require some additional testing with non-free word orders of course.

…And I think more and more that the ambiguity is desirable, because I now noticed that indeed, sentences like (3) are ambiguous, and I’ve been lacking that:

(3) Какой сон   видит Иван?
    which dream sees  Ivan
`What dream is Ivan seeing?' [rus]

The which here may indeed refer to both the subject (Ivan) and the object (dream).

To add to the above, I am currently not clear on what the MRS would look like, in terms of its ICONS, for, say, the two possible readings of sentence (2), one with focus on the determiner and one with focus on the whole NP.

Here’s what I am getting now, but I am not sure how to read this.
Screen Shot 2020-04-06 at 1.37.21 PM

There’s a lot here, but a few comments that might be helpful:

(1) I think in this case ambiguity is better than undergeneration.
(2) I don’t think you want the wh head-filler rule and the non-wh one being the same rule, unless the pattern really is identical AND you do something else to correlate [ SF ques ] with the presence of the wh word
(3) Focus on just the determiner v. the whole NP is vexed given current assumptions, because determiners don’t have their own ARG0.

1 Like

I see; but from the data point of view, would you agree that sentences like (5) are evidence that this is possible?

(5) Какой, ты  говоришь, человек приехал?
    Which  you say       person  arrived
    `Which person are you saying has arrived?' [rus]

So clearly the string is possible. As for what’s going on in the information structure — that requires a way to differentiate the information structural meaning of (5) compared to say (6) (assuming (6) is grammatical):

(6) Какой человек, ты говоришь, приехал?
Which person you say arrived
`Which person are you saying has arrived?’ [rus]

Yes, (6) is possible, and there will be intonational difference but I suppose I do not have a systematic understanding on how to test for information structure.

I don’t either. @sanghoun – do you have some wisdom for us here?