Preparing dev and test data

The AM parser only reads amconll files as input -- in order to give the dev or test data, we must create an amconll file that contains the sentences and further annotations like POS and lemmas, but the columns for the graph constants and the heads will be empty and are meant to be filled in by the parser.

ACL 2019

Dev Data

Test Data

EDS

• Download https://lindat.mff.cuni.cz/repository/xmlui/handle/11234/1-1956
• we only need train.amr.gz and test.amr.gz from the eds directory, unpack them.
• use PrepareTestData in eds.tools

MRP

Dev data

After having split the data, call this on the dev data:

java -cp am-tools-all.jar de.saar.coli.amrtagging.mrp.tools.PrepareDevData --mrp <MRP dev data.mrp> --companion <companion_data.conllu> --outPath <where to store> --prefix dev

The resulting file dev.amconll will contain no AM dependency trees but can serve as input to the parser.