This is the source code of the system of "Shallow Discourse Parsing Using Convolutional Neural Network" for CoNLL 2016 shared task, the focus is implicit discourse recognition using CNN.

The system is based on the best system (Wang and Lan, 2015) of CoNLL 2015, replacing the implicit part with CNN model.

• Note: some of the models (model_trainer/*/*.h5) are compressed with gzip to limit them below 100M, thus they should be decompressed before usage.