Deadline extension (sent 2017-05-12 10:06 CET)

Dear all,

based on the experience so far, we decided to extend the deadline for successful system runs by 48 hours (i.e. until mid-Monday in Europe). This comes at the cost of less time for the system description papers—our schedule is tight and we do not plan on extending the other deadlines. If you are done with your test runs, start writing the paper now and just add the results later.

Also, if you are desperate about the time needed for your system to complete, you may ask us to combine several partially successful runs of your system. Example: your run 1 has completed and the evaluator tells you that you failed to produce valid output for 10 of the 81 test files. You fix the bug but it would take too long to run the system again over all 81 files. So you temporarily modify the software to only process the 10 missing files, and perform run 2. While it is running, you have some time to add a tweak that will hopefully improve your performance on the surprise languages (for which you already have a valid output but you believe the accuracy must be quite bad). So you prepare the tweak and when run 2 successfully completes, you initiate run 3 that only processes the surprise languages. Then you ask us to take run 3 as the base; files missing in run 3 should be taken from run 2; and files missing in both should be taken from run 1. After run 3, do not forget to lift the temporary restriction and make sure that the system in the VM would process all the files if run anew!

Best regards,

Dan Zeman on behalf of the costocom :) (conll shared task organizing committee) http://universaldependencies.org/conll17/