Extracting an argument and an adjunct at the same time

Off to working with adjuncts (which I know should be tricky).

It’s my understanding that the following extracted-adjunct rule:

extracted-adv-adp-adj-phrase := basic-extracted-adj-phrase &
  [ SYNSEM [ LOCAL.CAT [ POSTHEAD #ph,
                         MC #mc ],
             NON-LOCAL.SLASH append-list & [ LIST < [ CAT [ HEAD +rp &
                                                       [ MOD < [ LOCAL intersective-mod &
                                                                       [ CAT [ HEAD #head,
                                                                               VAL #val,
                                                                               POSTHEAD #ph,
                                                                               MC #mc ],
                                                                         CONT.HOOK #hook,
                                                                         CTXT #ctxt ] ] > ],
                                                  VAL [ SUBJ olist,
                                                        COMPS olist,
                                                        SPR olist ] ] ]  > ] ],
    HEAD-DTR.SYNSEM canonical-synsem &
                    [ LOCAL local &
                            [ CAT [ HEAD verb & #head,
                                    VAL #val,
                                    POSTHEAD #ph,
                                    MC #mc ],
                              CONT.HOOK #hook,
                              CTXT #ctxt ],
                      NON-LOCAL.SLASH 0-alist,
                      MODIFIED notmod ],
    C-CONT [ HOOK #hook,
             RELS.LIST < >,
             HCONS.LIST <  >,
             ICONS.LIST <  > ] ].

will never take a slashed VP as a head daughter and furthremore produces a mother which has a SLASH list of strictly length 1.

I would ideally like to cover sentences like:

(1) Kto      kuda  idet?
     Who.NOM where goes
`Who is going where?' [rus]

(2) Ivan      kuda  idet?
    Ivan.NOM where goes
`Where is Ivan going?' [rus]

Now, I am aware of the fact that extracting adjuncts in an unrestricted way is supposed to be a can of worms, and I believe that (and I think even understand on some level some of the reasons).

Still, at the superficial syntax level, what would I do in order to relax the rule or potentially create a second rule to allow sentences like (1) and (2)?

I was able to get (1) by doing this:

extracted-adv-adp-adj-phrase := basic-extracted-adj-phrase &
  [ SYNSEM [ LOCAL.CAT [ POSTHEAD #ph,
                         MC #mc ],
             NON-LOCAL.SLASH append-list & [ LIST < [ CAT [ HEAD +rp &
                                                       [ MOD < [ LOCAL intersective-mod &
                                                                       [ CAT [ HEAD #head,
                                                                               VAL #val,
                                                                               POSTHEAD #ph,
                                                                               MC #mc ],
                                                                         CONT.HOOK #hook,
                                                                         CTXT #ctxt ] ] > ],
                                                  VAL [ SUBJ olist,
                                                        COMPS olist,
                                                        SPR olist ] ] ] . #slash > ] ],
    HEAD-DTR.SYNSEM canonical-synsem &
                    [ LOCAL local &
                            [ CAT [ HEAD verb & #head,
                                    VAL #val,
                                    POSTHEAD #ph,
                                    MC #mc ],
                              CONT.HOOK #hook,
                              CTXT #ctxt ],
                      NON-LOCAL.SLASH.LIST #slash,
                      MODIFIED notmod ],
    C-CONT [ HOOK #hook,
             RELS.LIST < >,
             HCONS.LIST <  >,
             ICONS.LIST <  > ] ].

But this adjunct extraction rule (while being potentially problematic in unrestricted application to itself), in combination with my current Subj-Head rule, will not cover (2) as it seems to insist that the mother’s MOD’s SUBJ is slashed (if I get this right).

I am confused because I can parse (3) without a problem where clearly there is also only one extraction, just like in (2):

(2) kuda  Ivan idet?
    where Ivan.NOM goes
`Where is Ivan going?' [rus]

Anyone sees my mistake(s)? (Again, let’s keep the question with the number of edges in the chart due to unrestricted application of ex-adj separate for now maybe).

What’s your evidence that (2) involves extraction? I.e. why isn’t that just an in-situ wh word?

Hm, well, I suppose we’ve decided that we are going to treat

(4) Kto      chto     vidit
     Who.NOM what.ACC sees
`Who sees what?'

as multiple extraction, so why would an adjunct be in-situ in this case of (1)? As for (2), why would it be in-situ at all, if it is the single wh-word?

This might be a good place to revisit this part of the analysis, actually.

Here’s the screenshot that I have from that meeting:

IMG_5358

It appears we said that it would make sense to analyze (4) as multiple extraction because this will help with ambiguity, somehow, but sadly I can’t reconstruct what exactly we were saying. I see that the second starred example (the one with the arrow) helps establish that some extraction must be involved but this does not have anything to do with multiple vs. subject-only extraction.

I guess evidence against subject only would be that the object can be at the front (and probably can also be the only thing at the front, though that’s perhaps a bit squishy):

(5) Chto     kto     vidit 
    what.ACC who.NOM sees  
`Who sees what?'

So, going back to (2), what should be some questions that I should be asking myself at this point, to analyze this further?

Hmm, I think I am starting to see; in (2), there is no obligatory fronting. And given that it probably requires special prosody, maybe it should be analyzed as echo. I don’t think it is echo though… But perhaps the prosody does indicate it may be out of scope for me, unless I want to reanalyze Russian as not having obligatory fronting (which is in fact a typological option). Or in any case, the prosody is an indication that this is a different kind of wh-word.

I don’t see where your extraction rule says anything about the mother having a MOD value. The MOD value is on the SLASHed constituent. I also don’t (yet) see what here would be causing (2) not to parse – it looks like a quite unconstrained rule to me. Maybe check with interactive unification?

I did, but yes, you are also right that I gave the wrong info, it is the SLASH-ed constituent’s MOD’s SUBJ’s SLASH that give the failure.

Ah, ok. I assume the analysis you are hoping for for (2) is extracting the adjunct from the V, then immediately filling it, then using the standard subj_hd rule? If so, which of those three rule applications is the unification failure showing up on?

I don’t think that’s a good analysis for this sentence. That is, in (2), if kuda is extracted, how are you expecting Ivan to attach? I think the head-filler rule expects to have a valence but gappy saturated daughter.

Yes, that’s right.

Now, the confusion above is due to the fact that I created those additional head-filler rules, and so now I have all those extra edges… I think I was looking at a head-2nd-filler rule (whose head daughter was the mother of the adjunct extraction) and was trying to unify it with the head daughter of subject-head.

Sorry about the confusion.

If I try to unify Subj-Head’s head daughter with the mother of the proper head-filler rule, then the failure is simply ARGS.FIRST.NON-LOCAL.SLASH.LIST (null vs. cons).

Indeed the VP that’s licensed by the head-filler says it’s subject has a nonempty SLASH list…

That is fishy, is it not? What does the next level down in the verbal projection say about the subject’s SLASH list? It’s hard to think how either the adjunct extraction rule or the filler head rule should be doing anything to the subject.

I don’t think you want a head-filler rule in this example. I would expect extracted things to go all the way to the left, not between the subject and the rest of the sentence. What is your evidence that there is extraction going on here, as opposed to wh-in-situ?

Ooh OK; my problem is also that I have currently wh-rules (head-filler) applying to head daughters which did not undergo extraction. So I was looking at a fishy edge altogether :/.

I am trying to get rid of these edges and not succeeding; I will post separately about it and then come back here. It is too confusing as it is.

See my response above; I am not sure but I was sort of trying to analyze it similarly to extracted subject-and-object…

I think the analysis you want is adj-head below subj-head.

With an extra lexical entry for when, then?

Yeah I see what you mean; this is of course different than extracted subj+obj as here the subject is not a wh-word and is not extracted.