AMR in AMR

[flipz357]

How would I best parse the sentence: "The cat runs is represented by the AMR "(r / run-01 :arg0 (c / cat))"...

Something like

(r / represent-01
     :arg0 (r / run-01
                     :arg0 (c / cat))
     :arg1 r
     :arg2 (m / meaning))

or

(r / represent-01
     :arg0 (r / run-01
                     :arg0 (c / cat))
     :arg1 (r2 / run-01
                     :arg0 (c2 / cat))
     :arg2 (m / meaning))

or rather

(r / represent-01
     :arg0 (s / string-entitiy
                     :op1 "(r / run-01 :arg0 (c / cat))")
     :arg1 (r2 / run-01
                     :arg0 (c2 / cat))
     :arg2 (m / meaning))

Maybe some version of the last approach is preferable?

[nschneid]

My gut feeling is that it's a string entity, because AMR notation is external to the language being annotated (English), or indeed, any natural language.

[flipz357]

Yes that was my gut feel as well, thx!

[flipz357]

@nschneid At least.... it's a fun challenge for a sota parser:

# This is a so-called abstract meaning representation for the sentence "The cat runs":  "(r / run-01 :arg0 (c / cat))"
(r / representation-02
   :ARG0 (t / this)
   :ARG1 (s / sentence
            :wiki -
            :name (n / name
                     :op1 "The"
                     :op2 "cat"
                     :op3 "run"))
   :mod (a / abstract)
   :mod (s2 / so-called)
   :ARG1-of (m / mean-01
               :ARG2 (s3 / string-entity
                         :value "avar0"
                         :ARG1-of (m2 / mean-01
                                      :ARG2 (s4 / string-entity
                                                :value "c")))))

(needed to sligthly adjust the example sentence since the other example sentence didn't work at all)

[jheinecke]

@nschneid, in case of taking

(r / represent-01
     :arg0 (s / string-entitiy
                     :op1 "(r / run-01 :arg0 (c / cat))")
     :arg1 (r2 / run-01
                     :arg0 (c2 / cat))
     :arg2 (m / meaning))

as the correct representation, are white spaces allowed in :op1 or should we split the string into

:op1 "(r/run-01" :arg0 (c / cat))")
:op2 ":arg0")
:op3 "(c/cat))"

or any other split or way to avoid spaces ?

The AMR guidelines are not explicit about spaces, but all examples in the guidelines (and in the LDC data) use :op1 ... :opN for strings instead of spaces

[nschneid]

Representing symbolic notation in AMR is uncharted territory AFAIK.

That said, the string-entity docs here only show an :op1 and it can contain spaces within the double quotes. As long as it's all on one line I think it should be fine.