Often we can guess from context whether amr-unknown refers to a person or not, as in these Dictionary examples:
What did the girl find? (find-01 :ARG0 girl :ARG1 amr-unknown)
Whose car is this? (car :poss amr-unknown :domain this)
But sometimes it is ambiguous: the first AMR would also apply to "Who did the girl find?" Likewise, currently "Who did you see?" and "What did you see?" both collapse to (see-01 :ARG0 you :ARG1 amr-unknown). AFAICT this is an exception in AMR: normally we do not remove animacy distinctions that are present in the language; "he" and "she" are retained rather than collapsed with "it", for example.
Would it be worth marking the distinction, e.g. mapping "who" to amr-unknown-person or amr-unknown :mod person ('which person')?
Often we can guess from context whether
amr-unknown
refers to a person or not, as in these Dictionary examples:(find-01 :ARG0 girl :ARG1 amr-unknown)
(car :poss amr-unknown :domain this)
But sometimes it is ambiguous: the first AMR would also apply to "Who did the girl find?" Likewise, currently "Who did you see?" and "What did you see?" both collapse to
(see-01 :ARG0 you :ARG1 amr-unknown)
. AFAICT this is an exception in AMR: normally we do not remove animacy distinctions that are present in the language; "he" and "she" are retained rather than collapsed with "it", for example.Would it be worth marking the distinction, e.g. mapping "who" to
amr-unknown-person
oramr-unknown :mod person
('which person')?