LineBert

This repository contains the code associated to the LineBert paper. It indroduces a simple method for discourse parsing in dialogue.

Scores	Attachments	Relations
STAC	73.06	56.25
STAC-squished	79.93	71.22

Data sets

The two data sets, stac and stac_squished can be found here

Setup

All the requirement for LineBert can be found in requirements.txt. For training or testing, the data sets should be put in the corresponding folder.
For training, run the following commands:

python bert_finetune.py
python linear.py
python multitask.py

• For evaluation add the trained models to the models folder, or use the .pth obtained after training, and run the the notebooks bert_finetune_test.ipynb and multitask_test.ipynb.