Open danielhers opened 7 years ago
Hi Daniel, | Are the PropBank frame files out of date, or does AMR use a different set of frames, as opposed to what is said in the README?
AMR uses what amounts to a "fork" of our frames. Since AMR and Propbank are both annotating new data in parallel and keep adding things, they need to be actively kept in sync. We've been working to make sure that any revisions that happen on the AMR side get into our frames, to that it's close to being a subset of ours, but some things can slip through ( and I hadn't yet run any automatic checks on the new BioAMR release.)
| Specifically, ablate.01 has only roles 0, 1 and 2, but in AMR it also has role number 3.
Thanks for noting that! That was a revision to the AMR frames that slipped through our update process; I'm going to run checks on BioAMR and make sure nothing else slipped through, and then make sure these get into the public frames.
(We hope to eventually have a shared set of frames used by both Propbank and AMR, with usage flags for AMR-only and Propbank-only rolesets and roles.)
Thanks a lot! That answers my question. Should I close the issue or keep it open until the data is synced?
Another difference: play.11
has ARG3: data carrier (score, record, tape, CD, DVD)
in AMR but not in PropBank.
And raise.02
has ARG3: recipient of money (if different from arg0)
.
I am aware of the fact that AMR has special rolesets ending in
-91
, but I noticed there is also at least one difference in "normal" rolesets. Specifically,ablate.01
has only roles 0, 1 and 2, but in AMR it also has role number 3. This role is used, for example, in AMRbel_pmid_1222_6756.17894
in the BioAMR data set:Are the PropBank frame files out of date, or does AMR use a different set of frames, as opposed to what is said in the README?