Structured Dialogue Discourse Parsing

Pytorch implementation of the paper Structured Dialogue Discourse Parsing.

Installation

Run the following commands to recreate the environment we used for the development of this project:

conda create -n sddp python=3.9.6 -y
conda activate sddp
conda install pytorch==1.12.1 cudatoolkit=11.3 -c pytorch -y
pip install -r requirements.txt
pip install git+https://github.com/timvieira/arsenal.git@b7a5f93b04533236bbe297b4079256ea15812b88
pip install -e git+https://github.com/rycolab/spanningtrees.git@68c68e6f37d0951a524b901d9e84c5c1c498420e#egg=spanningtrees

Data Preparation

Please pass --link_only to all the shell scripts if you want to train and test using the link_only option. The default is link and relation.

cd data/
bash parse.sh (--link_only)
cd ../

Train/save on one domain

bash train_stac.sh (--link_only)
bash train_molweni.sh (--link_only)

Test on one domain

bash test_stac.sh (--link_only)
bash test_molweni.sh (--link_only)

Train on one domain and test/save on another domain

bash train_stac_on_molweni.sh (--link_only)
bash train_molweni_on_stac.sh (--link_only)

Test on another domain

bash test_stac_on_molweni.sh (--link_only)
bash test_molweni_on_stac.sh (--link_only)