[epe-users] EPE 2018: final results now publicly available

[Stephan Oepen]

dear colleagues,

we have just released through the task web page the final and official
results for the EPE 2018 add-on exercise to the 2018 UD Parsing Shared
Task.  please see:

https://goo.gl/3Fmjke

we are sorry that it took us a little more than two extra weeks since
we had posted initial scores to arrive at this point.  in a nutshell,
there were a couple of submissions that required debugging (and
increased robustness to whitespace variation) in the opinion system;
and we fixed a bug in the negation system and implemented automated
hyper-parameter tuning for it (which the other two systems already had
been doing).

we have started to try and make sense of these results and, in
particular, are lookin for correlations to various intrinsic
evaluation metrics.  we will contribute an EPE overview paper to the
proceedings for the UD Parsing Shared Task, and we do of course look
forward to discussing EPE 2018 results and the larger picture later
this fall in brussels.

in case you discuss EPE performance in your system description, please
make sure to use the latest, official numbers.  a BibTeX entry for the
overview paper is available for your convenience as part of a
collection of EPE bibliographic references:

http://svn.nlpl.eu/epe/2018/public/epe.bib

warmest thanks, once more, to everyone who contributed!  we believe
there will be several interesting observations coming out of this, and
we will share a draft of the overview paper well before the
camera-ready deadline for system descriptions.  in case you notice
anything surprising or have general feedback or suggestions for
improvement, please do not hesitate to get in touch with us!

best wishes, oe (for the EPE 2018 co-organizers)