Difficult to avoid multiple NEs in the same sentence

[mgeorgescu]

Japan has been expanding its space programme and is now conducting the most extensive probe of the Moon since the United States Apollo missions of the decades of 600000 and 700000 .

From:

(a / and
      :op1 (e / expand-01
            :ARG0 (c / country :name (n / name :op1 "Japan"))
            :ARG1 (p / program
                  :poss c
                  :mod (s / space)))
      :op2 (c2 / conduct-01
            :ARG0 c
            :ARG1 (p2 / probe-01
                  :ARG1 (m4 / moon :name (n2 / name :op1 "Moon"))
                  :time (s3 / since
                        :op1 (m2 / mission :name (n3 / name :op1 "Apollo")
                              :mod (c3 / country :name (n4 / name :op1 "United" :op2 "States"))
                              :time (d / date-entity :decade 1960))
                        :op2 (m3 / mission :name (n6 / name :op1 "Apollo")
                              :mod c3
                              :time (d2 / date-entity :decade 1970)))
                  :mod (e2 / extensive
                        :degree (m / most)))
            :time (n5 / now)))

To

(a / and
      :op1 (e / expand-01
            :ARG0 (c / country :name (n / name :op1 "Japan"))
            :ARG1 (p / program
                  :poss c
                  :mod (s / space)))
      :op2 (c2 / conduct-01
            :ARG0 c
            :ARG1 (p2 / probe-01
                  :ARG1 (m4 / moon :name (n2 / name :op1 "Moon"))
                  :time (s3 / since
                        :op1 (m2 / mission :name (n3 / name :op1 "Apollo")
                              :mod (c3 / country :name (n4 / name :op1 "United" :op2 "States"))
                              :time (d / date-entity :decade 1960))
                        :op2 (m3 / mission :name n2
                              :mod c3
                              :time (d2 / date-entity :decade 1970)))
                  :mod (e2 / extensive
                        :degree (m / most)))
            :time (n5 / now)))

However, this is picked up as incorrect by the checker (:name n2). Coreference cannot be used as the 2nd instance of the NE is qualified.

Another similar example is:

In June 2004 Hong Kong Polytechnic University (PolyU) joined national university science and technology park of the Virtual University Park in Shenzhen, China becoming the first tertiary institution of Hong Kong to establish research centers and training facilities in mainland China.

(j / join-01
      :ARG0 (u / university :name (n / name :op1 "Hong" :op2 "Kong" :op3 "Polytechnic" :op4 "University"))
      :ARG1 (p / park
            :mod (u2 / university)
            :poss (e / event :name (n3 / name :op1 "Virtual" :op2 "University" :op3 "Park")
                  :location (c / city :name (n4 / name :op1 "Shenzhen")
                        :location (c2 / country :name (n5 / name :op1 "China"))))
            :mod (s / science)
            :mod (n2 / national)
            :mod (t / technology))
      :time (d / date-entity :year 2004 :month 6)
      :mod (b / become-01
            :ARG1 u
            :ARG2 (i / institute
                  :mod (t2 / tertiary)
                  :ord (o / ordinal-entity :value 1)
                  :poss (c3 / city :name (n6 / name :op1 "Hong" :op2 "Kong"))
                  :ARG0-of (e2 / establish-01
                        :ARG1 (a2 / and
                              :op1 (c4 / center
                                    :mod (r / research-01))
                              :op2 (f / facility
                                    :mod (t3 / train-01)))
                        :location (c5 / country :name (n7 / name :op1 "China")
                              :mod (m / mainland))))))

OR:

Sentence: Saeed Jalili's party and Javier Solana's party stated that Iran's chief nuclear negotiator Saeed Jalili and European unuion foreign policy chief Javier Solana plan to hold talks in Brussels on 080123 to discuss Iran's nuclear program.

(s / state-01
      :ARG0 (a / and
            :op1 (p5 / party
                  :ARG1-of (h4 / have-org-role-91
                        :ARG0 p2))
            :op2 (p6 / party
                  :ARG1-of (h5 / have-org-role-91
                        :ARG0 p9)))
      :ARG1 (p7 / plan-01
            :ARG0 (a2 / and
                  :op1 (p2 / person :name (n5 / name :op1 "Saeed" :op2 "Jalili")
                        :ARG0-of (h2 / have-org-role-91
                              :ARG1 (c4 / country :name (n8 / name :op1 "Iran"))
                              :ARG2 (c5 / chief)
                              :ARG3 (n9 / negotiate-01
                                    :ARG2 (n10 / nucleus))))
                  :op2 (p9 / person :name (n6 / name :op1 "Javier" :op2 "Solana")
                        :ARG0-of (h3 / have-org-role-91
                              :ARG1 (o / organization :name (n3 / name :op1 "European" :op2 "Union"))
                              :ARG2 (c / chief)
                              :ARG3 (p8 / policy
                                    :mod (f / foreign)))))
            :ARG1 (h / hold-04
                  :ARG1 (t / talk-01
                        :purpose (d2 / discuss-01
                              :ARG1 (p3 / program-01
                                    :mod (n2 / nucleus)
                                    :mod c4))
                        :location (c3 / city :name (n7 / name :op1 "Brussels"))
                        :time (d / date-entity :day 23 :month 1 :year 2008)))))

[mgeorgescu]

One other tricky case is the following, where it's tricky to handle the 3rd instance of "Australia".

After strategically ' excising ' bits of Australia in order to thwart people smugglers ( you can n't land in Australia if we pass a law to say it IS N'T Australia this week ), our pint @-@ sized PM has decided to think big .

(m / multi-sentence
     :snt1 (d / decide-01
          :ARG0 (p / person
               :ARG0-of (h / have-org-role-91
                    :ARG2 (m2 / minister
                         :mod (p2 / prime)))
               :ARG1-of (s / size-01
                    :ARG2 (p3 / pint))
               :poss (w / we))
          :ARG1 (t / think-01
               :ARG1 p
               :ARG2 (b / big))
          :time (a / after
               :op1 (e / excise-01
                    :ARG0 p
                    :ARG1 (b2 / bit
                         :part-of (c2 / country
                              :name (n3 / name :op1 "Australia")))
                    :manner (s2 / strategic)
                    :purpose (t2 / thwart-01
                         :ARG0 p
                         :ARG1 (p4 / person
                              :ARG0-of (s3 / smuggle-01
                                   :ARG1 (p7 / person)))
                         :ARG0-of (m3 / mean-01
                              :ARG1 (p5 / possible
                                   :polarity -
                                   :domain (l / land-01
                                        :ARG1 (y / you)
                                        :location c2)
                                   :condition (p6 / pass-01
                                        :ARG0 (w3 / we)
                                        :ARG1 (l2 / law
                                             :ARG0-of (s4 / say-01
                                                  :ARG1 (c4 / country
                                                       :polarity -
                                                       :name (n5 / name :op1 "Australia")
                                                       :time (d2 / date-entity
                                                            :unit (w2 / week
                                                                 :mod (t3 / this)))
                                                       :domain c2))))))))))) 

Ulf suggested using have-part-01 (reification of :part) in last occurrence, instead of :domain. However, like this we feel that we would lose the idea of "Australia is not Australia this week".

If you have any ideas on how to better represent this, please feel free to suggest. Thank you

[nschneid]

Love the Australia sentence. :) I think "pass a law to say it isn't Australia this week" means that the name "Australia" is being negated. I don't see an official reification of :name, but this could be represented with something like

(s4 / say-01
   :ARG1 (n6 / name-01
             :polarity -
             :ARG1 c2
             :ARG2 (n5 / name :op1 "Australia"))
   :time (d2 / date-entity
             :unit (w2 / week
                       :mod (t3 / this))))

(I read "this week" as modifying "say", though it is debatable.)

[uhermjakob]

Case 1: Apollo missions. I don't think that there were two separate series of Apollo missions, so we can just use a date-interval in this case:

(m / mission :name (n / name :op1 "Apollo")
  :mod (c / country :name (n2 / name :op1 "United" :op2 "States"))
  :time (d / date-interval
          :op1 (d2 / date-entity :decade 1960)
          :op2 (d3 / date-entity :decade 1970)))

(Also, since should generally only have an :op1 and no additional :op2.)

Case 2: China and mainland China Sometimes there are two legitimately different entities. Obvious case: Paris, France vs. Paris, Texas. This case is a bit special as one is part of the other. Following our precedent for They live in the south of France. which we annotate as south :part-of France, we might want to the same for mainland China. So I propose:

(c / country :name (n / name :op1 "China"))
...
(m / mainland
  :part-of c)

Case 3: Saeed Jalili, Javier Solana, Iran I'm not sure what the problem is. The three NEs above play multiple roles, which the AMR properly describes.

Case 4: Australian migration zone Super tricky sentence. I believe that the it actually does not refer to Australia, but to the location where boat refugees land. Background: (1) Australia generally allows refugees to ask for asylum. (2) Australia does not want too many refugees coming by boat from Indonesia. Solution: Define an "Australian migration zone" where boat people may ask for asylum, but exclude from this migration zone some islands such as Christmas Island and other islands that are easy to reach by boat from Indonesia, so that Christmas Island is not considered part of Australia for asylum purposes. I think that the it refers to these islands excised from the "Australian migration zone". If you land there, it does not count as Australia for asylum purposes. So I would annotate " it isn't Australia" as " it (the little islands where boat refugees land) is not part of Australia (for asylum purposes)."

By the way, in an unprecedented move, Australia recently excised mainland Australia from the Australian migration zone, so that now 99.9% of Australia does not count as Australia for asylum purposes. (That's probably what the think big at the end of the sentence refers to.)

[cbonial]

I've also experienced a problem with sentences such as the sentence below, which mentions North Korea as a whole and then "northeastern North Korea" specifically.

Should we introduce a generic NE here and say that it is "part-of" North Korea? As a side note, this issue touches on some of the overall coreference issues that come up when an entity changes over the time span of the utterance, but AMR treats this statically.

[uhermjakob]

Based on the following example from our guidelines: "They live in the south of France."

(l / live-01
  :ARG0 (t / they)
  :location (s / south
              :part-of (c / country :name (n / name :op1 "France"))))

I would annotate "... North Korea ... northeastern North Korea ..." as

(c / country :name (n / name :op1 "North" :op2 "Korea"))
...
(n2 / northeast
   :part-of c)

[mgeorgescu]

Hi all, Thank you for your feedback.

Ulf, special thanks for the in depth explanation about Australia. We updated the sentences accordingly.