poss-phrase-1 missing binary-nonloc-phrase supertype

[emilymbender]

Another spot where the adnominal possession library leaves the NON-LOCAL features underspecified. poss-phrase-1 should have binary-nonloc-phrase as a supertype here:

poss-phrase-1 := head-final &
  [ SYNSEM.LOCAL.CAT [ HEAD #head,
                       VAL [ COMPS < >,
                             SUBJ < >,
                             SPEC < >,
                             SPR < > ] ],
    HEAD-DTR.SYNSEM.LOCAL [ CAT [ POSSESSUM nonpossessive,
                                  HEAD #head & noun &
                                       [ POSSESSOR nonpossessive,
                                         PRON - ],
                                  VAL.SPR < [ ] > ],
...

Also, I have a choices file (with lots more info in) where the resulting grammar lacks a rule instantiating poss-phrase-1. Couldn't reproduce it with this minimal example though, so I'll store that choices file locally. [Note to self: it will be in the same email thread as this.]

choices.txt

[emilymbender]

Edit: the larger choices file does seem to have produced the rule. So, the only problem is the one reproduced in the smaller choices file.

[ltxom]

I'm trying to add a regression test for this. Can you please give me an example of a poss-phrase passing up the NON-LOCAL features in binary phrases?

[emilymbender]

I don't have time to construct the example right now, but I think what is needed is adding choices to the choices file that describe in situ wh questions. This will cause the wh-in-situ rule to be instantiated, and then if non-local features are underspecified, it will probably fire in every sentence.

[olzama]

Let me know if you still need help with this, @ltxom

[ltxom]

@olzama Thank you for your reply! Could you please construct me an example of how underspecified non-local features fire in every sentence with instantiated wh-in-situ rule?

[olzama]

Actually, I am myself unable to reproduce the problem based on the given description.

I did:

(1) Download the attached choices file (it's a version of the "minimal grammar" but with some adnominal possession choices (2) Add a choice for in situ wh-questions (3) Add a question pronoun (4) Save the changes (5) Download the updated grammar (6) Loaded it into LKB-FOS (7) Tried parsing a sentence n1 iv and then also who iv

But I didn't get the spurious in-situ parse for n1 iv.

@emilymbender do you remember more details? (I know the issue is widespread in general but in this case I am not sure which exact part of it we are talking about.)

[emilymbender]

In that grammar, what does poss-phrase-1 look like? Does the grammar have an in-situ phrase?

[olzama]

Yes, it does have the in-situ phrase. I attach the choices.

poss-phrase-1 := head-final &
  [ SYNSEM.LOCAL.CAT [ HEAD #head,
                       VAL [ COMPS < >,
                             SUBJ < >,
                             SPEC < >,
                             SPR < > ] ],
    HEAD-DTR.SYNSEM.LOCAL [ CAT [ POSSESSUM nonpossessive,
                                  HEAD #head & noun &
                                       [ POSSESSOR nonpossessive,
                                         PRON - ],
                                  VAL.SPR < [ ] > ],
                            CONT.HOOK #hook &
                                      [ LTOP #lbl,
                                        INDEX #possessum &
                                              [ COG-ST uniq-id ] ] ],
    C-CONT [ HOOK #hook &
                  [ INDEX #possessum ],
             ICONS.LIST < >,
             RELS.LIST < arg12-ev-relation &
                         [ PRED "poss_rel",
                           LBL #lbl,
                           ARG1 #possessum,
                           ARG2 #possessor ],
                         quant-relation &
                         [ PRED "exist_q_rel",
                           ARG0 #possessum,
                           RSTR #harg ] >,
             HCONS.LIST < qeq &
                          [ HARG #harg,
                            LARG #lbl ] > ],
    NON-HEAD-DTR.SYNSEM.LOCAL [ CONT.HOOK.INDEX #possessor,
                                CAT [ POSSESSUM nonpossessive,
                                      VAL.SPR < >,
                                      HEAD noun &
                                           [ POSSESSOR nonpossessive,
                                             CASE obl,
                                             PRON - ] ] ] ].

choices.txt

[emilymbender]

Ah, the thing is that the poss-phrase-1 has to be in the sentence for the problem to show up. So, try: "n1 n1 iv".

[olzama]

Ah, I see, thank you, Emily.

OK @ltxom , then your minimal example is:

1) Use the choices file obtained as outlined here: https://github.com/delph-in/matrix/issues/632#issuecomment-1117163124 2) Use the string n1 n1 iv as the test sentence (test suite). 3) The desired result is one tree, with underspecified SF value in the MRS 4) The bad result would be two trees, of which one is a question [SF ques]

Let me know if you have questions.

[ltxom]

Thank you so much, @olzama @emilymbender ! I could reproduce the example and see the bad result with two trees. I will investigate more on the libraries and come up with solutions.

[emilymbender]

@Diana-BZ This one would be a good target for fixing! (I just hit it again in 567...)

[Diana-BZ]

I'm trying to add a regression test for this. Can you please give me an example of a poss-phrase passing up the NON-LOCAL features in binary phrases?

Hi @ltxom! Is there already a regression test added for this issue? If so, do you know what the name of it is?