srl_v*.py - Extracts SRL features in CRFsuite format
Get training and test data for German. You also need CRFsuite installed.
Specify the features you want to use in the source code (ugly, I know). Run srl_v*.py on test and training data to get output files in CRFsuite format. Note that in the paper, the development set is used as test set.
$ python srl_v*.py <data file>
Learn a CRF model from the output of the training data.
$ crfsuite learn -m <model name> <training data file>
Test the model on the test data.
$ crfsuite tag -qt -m <model name> <test data file>
Previous results on the models descriped in the paper can be found in res.all.txt.
The task is to identify semantic arguments of sentence predicates and classify them into their semantic roles.
For example:
[ John ]AGENT hits [ Mary ]PATIENT [ with a stick ]INSTRUMENT .
Read the paper. Find more details on the original CoNLL-2009 Shared Task.
Dependency representations following the SRL CoNLL-2009 Shared Task.
Melanie Tosik, tosik@uni-potsdam.de