Splitting a single dialogue task into multiple tasks with dropping different output tokens to learn better encoder representation.
An arxiv link to the paper can be found here.
python beamsearch2.py
for training and reproducing the results.
@inproceedings{
parthasarathi2021Semantic,
author = {Parthasarathi, Prasanna and Abdelsalam, Mohamed and Pineau, Joelle and Chandar, Sarath},
title = {A Brief Study on the Effects of Training Generative Dialogue Models with a Semantic loss},
year = {2021},
booktitle = {Proceedings of the 22nd Annual SIGdial Meeting on Discourse and Dialogue},
publisher = {Association for Computational Linguistics},
}