[mrp-users] final (we hope) updates to MRP 2020 training and companion data

[Stephan Oepen]

dear colleagues:

yesterday, we handed over another update of the MRP 2020 data package
to the LDC, where it is now available for no-cost download to
registered task participants.  the latest version of our training,
validation, and companion data makes some (relatively minor)
improvements in the chinese AMR and german DRG annotations, adds
chinese companion dependency parses and the english parses for some
sentences missing in the original release, and provides reference
anchorings (aka alignments in AMR parsing) for the english AMR and
english and german DRG annotations.  as with earlier updates, this new
data package supersedes all previous releases, i.e. please make sure
to download the latest version from the LDC.  to validate you have the
right version: the various README files included in the package should
indicate a release date of june 22, 2020.

preparing all the companion data has taken more time than we had
anticipated, and we have decided to extend the duration of the shared
task by one week to compensate for the delay in final data
availability.  please note that the evaluation period is now scheduled
to run from july 27 through august 10, 2020.  we will be back with
more specific instructions in mid-july on how to submit your parser
outputs for evaluation, but in a nutshell parser inputs will be
distributed (on july 27) as MRP files with the ‘tops’, ‘nodes’, and
‘edges’ fields missing, and we will ask that you upload (by august 10)
an enhanced version, again in MRP format, with the actual graphs
filled in.

as you may have seen, we have just circulated a final call for
participation to various mailing lists and have sought to generate
some visibility on twitter.  we would be grateful if you could assist
in spreading the word, for example linking to the task web site
(http://mrp.nlpl.eu) from your personal or institutional web pages,
creating more ‘buzz’ on social media, and maybe most importantly
encouraging colleagues who work in meaning representation parsing to
consider making a submission.  as we have tried to clarify in the most
recent call for participation, partial submissions (in the sense of
only targeting a subset of frameworks and languages) are possible and
will be welcome, in part simply to help snapshot the current state of
the art.  in other words, anyone with a functional parser for at least
one of our five frameworks should ideally make a submission, if maybe
only putting moderate effort into training from and serializing parser
outputs to the uniform MRP serialization.

finally, as always: please do not hesitate to get in touch with us (at
the ‘mrp-organizers’ address) to ask any open questions you might have
or make suggestions for extending or adjusting the task design!  at
this point, i believe it is fair to say that the MRP 2020 shared task
is about to enter its ‘hot’ phase: a little less than seven weeks
remain until the closing date of the evaluation phase :-).

best wishes, oe (for the MRP 2020 co-organizers)