amrisi / amr-guidelines

246 stars 86 forks source link

Multiple choice questions #185

Open nschneid opened 8 years ago

nschneid commented 8 years ago

What did we decide to do with "multiple choice" questions like "Do you want food or (do you want) drink?". Couldn't find any documentation here or in the AMR Dictionary.

kevincrawfordknight commented 8 years ago

Y-N-QUESTION versus PICK-ONE-OR-MORE-QUESTION ("do you want/prefer coffee or tea?").

The latter is paraphrased as "what do you want? (coffee, tea, both, neither)".

(w / want-01
  :arg0 (y / you)
  :arg1 (o / AMR-CHOICE                  ?????
          :op1 (c / coffee)
          :op2 (t / tea)))

Ulf: Don't need :mode interrogative here. Tim: Kind of strange to have interrogative for "whether/if".

(w / want-01
  :arg0 (y / you)
  :arg1 (o / or
          :op1 (c / coffee)
          :op2 (t / tea))
  :mode choice)                          ?????

Nathan: Doesn't specify what the choices are.

nschneid commented 8 years ago

I think we agreed on the amr-choice solution. There are 357 release AMRs with :mode interrogative on sentences containing "or", so presumably these would need to be inspected during retrofitting.

timjogorman commented 7 years ago

Kira mentioned the issues of using amr-choice for "I don't know whether he wants coffee or tea". I'd like to propose a simplistic alternative just using include-91:

"Do you want coffee or tea?"

(w / want-01
      :ARG1 (a2 / amr-unknown
            :ARG1-of (i3 / include-91
                  :ARG2 (o3 / or
                        :op1 (t / tea)
                        :op2 (c / coffee)))))
(what do you want from the set of tea and coffee)

"Should I stay or should I go?"

(o / obligate-01
      :ARG1 (a / amr-unknown
            :ARG1-of (i2 / include-91
                  :ARG2 (o2 / or
                        :op1 (s / stay-01
                              :ARG1 (i / i))
                        :op2 (g / go-02
                              :ARG0 i)))))

"I don't know whether he wants coffee or tea"

(k / know :polarity -
      :ARG0 i
      :ARG1 (t2 / thing
            :ARG1-of (w2 / want-01
                  :ARG0 (h / he))
            :ARG2-of (i4 / include-91
                  :ARG1 (o4 / or
                        :op1 (t3 / tea)
                        :op2 (c2 / coffee)))))
nschneid commented 7 years ago

Maybe, to parallel the truth-value treatment in #193:

"I don't know whether he wants coffee or tea"

(k / know :polarity -
      :ARG0 i
      :ARG1 (t2 / thing
            :ARG1-of (w2 / want-01
                  :ARG0 (h / he))
            :ARG2-of (c / choice-value
                  :op1 (t3 / tea)
                  :op2 (c2 / coffee))))
nschneid commented 7 years ago

Observation: truth-value in #193 has an implicit 'or not' in its meaning:

"I don't know whether he wants coffee" = choice(wants-coffee, not(wants-coffee)) "I don't know whether he wants coffee or tea" = choice(wants-coffee, wants-tea)

timjogorman commented 7 years ago

I'd be fine with choice-value instead of include-91 there. My main thought is just to split the "amr-unknown" from whatever predicate we use to describe the range of choices (which, as Kira pointed out, is sometimes not a question) -- I'm fine with having a special concept or roleset for that "range of choices".