Issue: For a phrase such as long-range, both concepts now have roles for each other:
range-01 :ARG2 long-03 ("What's the extent of the range?)
long-03 :ARG1 range-01 ("What is long?")
Potential consequences: (1) a cyclic AMR and (2) graph theorists freak out
Thanks, Kira, for raising this super interesting issue.
I agree with you to try to avoid creating cycles, at least for the upcoming release and until further discussion with the larger AMR group. So I support your solution:
(r / range-01
:ARG2 (l / long-03)))
More generally, when two concepts have roles for each other, events would take precedence over objects, which in turn would take precedence over properties (such as long-03).
Issue: For a phrase such as long-range, both concepts now have roles for each other:
Thanks, Kira, for raising this super interesting issue.
I agree with you to try to avoid creating cycles, at least for the upcoming release and until further discussion with the larger AMR group. So I support your solution:
More generally, when two concepts have roles for each other, events would take precedence over objects, which in turn would take precedence over properties (such as long-03).
Alternative for long-range missile with cycle:
But let's revisit this at one of the next AMR meetings, both for its theoretical and pragmatic implications.
On the other hand, annotators have been adding "long-03 :ARG1 ..." in cases where it appears reasonable and does not lead to cycles as in:
"The meeting lasted very long."