coordinated modal ellipsis with wh argument

nschneid commented 5 years ago

@mdtux89 raises this interesting example: "Who can afford it and who can't?"

Option A:

(a / and
      :op1 (p / possible-01
            :ARG1 (a2 / afford-01
                  :ARG0 (a3 / amr-unknown)
                  :ARG1 (i2 / it)))
      :op2 (p2 / possible-01 :polarity -
            :ARG1 a2))

Option B:

(a / and
      :op1 (p / possible-01
            :ARG1 (a2 / afford-01
                  :ARG0 (a3 / amr-unknown)
                  :ARG1 (i2 / it)))
      :op2 (p2 / possible-01 :polarity -
            :ARG1 (a4 / afford-01
                  :ARG0 (a5 / amr-unknown)
                  :ARG1 i2)))

The rationale for Option B is that the two "who"s refer to distinct sets of people, the haves and the have-nots, so perhaps this should be explicit with separate amr-unknowns.

A similar rationale would apply if the modals are removed, e.g. "Who likes fish and who doesn't?". Except then it is clearer that the main event ("likes") needs to be duplicated in order to host the negation in the second instance.

mdtux89 commented 5 years ago

There is an instance of this in the LDC corpus, where they seem to follow Option B:

"Not knowing how much profit is in there, not caring who can afford it and who can't, these do not seem to be the principles of socialism."

(s / seem-01
      :ARG1 (p4 / principle :polarity -
            :poss (s2 / socialism)
            :domain (a / and
                  :op1 (k / know-01 :polarity -
                        :ARG1 (t2 / thing
                              :quant-of (p / profit
                                    :location (t / there))))
                  :op2 (c / care-01 :polarity -
                        :ARG1 (a3 / and
                              :op1 (p5 / person
                                    :ARG0-of (a4 / afford-01
                                          :ARG1 (i2 / it)
                                          :ARG1-of (p2 / possible-01)))
                              :op2 (p6 / person
                                    :ARG0-of (a6 / afford-01
                                          :ARG1 i2
                                          :ARG1-of (p3 / possible-01 :polarity -))))))))

I just wasn't sure if this was following some guidelines or an individual annotator decision.

luciaelizabeth commented 5 years ago

Is something like what is below possible? The “affording” refers to the same hypothetical event, but the participant is different.

Lucia

Option C:

(a / and

:op1 (p / possible-01

:ARG1 (a2 / afford-01

:ARG0 (a3 / amr-unknown)

:ARG1 (i2 / it)))

:op2 (p2 / possible-01 :polarity -

:ARG1 a2

:ARG0 (a5 / amr-unknown)

:ARG1 i2)))

From: Nathan Schneider notifications@github.com Reply-To: amrisi/amr-guidelines reply@reply.github.com Date: Thursday, 4. July 2019 at 16:33 To: amrisi/amr-guidelines amr-guidelines@noreply.github.com Cc: Subscribed subscribed@noreply.github.com Subject: [amrisi/amr-guidelines] coordinated modal ellipsis with wh argument (#240)

Option A:

(a / and :op1 (p / possible-01 :ARG1 (a2 / afford-01 :ARG0 (a3 / amr-unknown) :ARG1 (i2 / it))) :op2 (p2 / possible-01 :polarity - :ARG1 a2)) Option B:

(a / and :op1 (p / possible-01 :ARG1 (a2 / afford-01 :ARG0 (a3 / amr-unknown) :ARG1 (i2 / it))) :op2 (p2 / possible-01 :polarity - :ARG1 (a4 / afford-01 :ARG0 (a5 / amr-unknown) :ARG1 i2)))

nschneid commented 5 years ago

Option C:

(a / and
       :op1 (p / possible-01
             :ARG1 (a2 / afford-01
                   :ARG0 (a3 / amr-unknown)
                   :ARG1 (i2 / it)))
       :op2 (p2 / possible-01 :polarity -
             :ARG1 a2
                   :ARG0 (a5 / amr-unknown)
                   :ARG1 i2)))

No, since each variable refers to a unique node in the graph, in Penman notation it doesn't make sense to repeat a variable with different things underneath.

luciaelizabeth commented 5 years ago

Ah, correct. I don't think AMR uses the able-01 frame from Propbank ( http://verbs.colorado.edu/propbank/framesets-english-aliases/able.html), but this sentence seems to read more like the following to me than the other options:

(a / and :op1 (a / able-01 :ARG1 (a2 / amr-unknown :ARG2 (a3 / afford-01 :ARG1 (i / it))) :op2 (a4 / able-01 :polarity - :ARG1 (a5 / air-unknown) :ARG2 a3))

Lucia Donatelli http://luciadonatelli.georgetown.domains/

On Thu, Jul 4, 2019 at 5:46 PM Nathan Schneider notifications@github.com wrote:

Option C:

(a / and :op1 (p / possible-01 :ARG1 (a2 / afford-01 :ARG0 (a3 / amr-unknown) :ARG1 (i2 / it))) :op2 (p2 / possible-01 :polarity - :ARG1 a2 :ARG0 (a5 / amr-unknown) :ARG1 i2)))

No, since each variable refers to a unique node in the graph, in Penman notation it doesn't make sense to repeat a variable with different things underneath.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/amrisi/amr-guidelines/issues/240?email_source=notifications&email_token=AGGDFSFS3EDVBDHH57PC6HDP5YLO7A5CNFSM4H53UKU2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODZHXNLQ#issuecomment-508524206, or mute the thread https://github.com/notifications/unsubscribe-auth/AGGDFSGYFPV7FMYO7RAQP2LP5YLO7ANCNFSM4H53UKUQ .

nschneid commented 5 years ago

I don't think AMR uses the able-01 frame from Propbank

Correct. And if it did, the ARG1 of able-01 would probably be marked as coreferent with the ARG0 of afford-01 (as a case of control), so we'd face the same problem of whether to duplicate afford-01.

amrisi / amr-guidelines

coordinated modal ellipsis with wh argument #240