Closed luheng closed 7 years ago
Thanks for your attention, the network architecture in demo is consistent with that paper, while the parameters setup and features are not the optimal. Currently, it's provided to users for study. The experiment code of paper was implemented on the past interface of PaddlePaddle, which is different from this demo. We will update the setup later, after verified experiment on new code.
Thank you so much for your timely reply!
I'm wondering if you happen to have test results from training the current SRL code on the CoNLL2005 training set? And if so, are they close to the numbers reported in the ACL 2015 paper?
@luheng We update SRL demo recently. Now, it is same as 'End-to-end Learning of Semantic Role Labeling Using Recurrent Neural Networks'
Is the setup in demo/semantic_role_labeling/train.sh a full replication of the ACL 2015 paper End-to-end Learning of Semantic Role Labeling Using Recurrent Neural Networks? Thanks!