Skip to content

Code and data for the paper "Facilitating the Communication of Politeness through Fine-Grained Paraphrasing". EMNLP 2020.

Notifications You must be signed in to change notification settings

CornellNLP/politeness-paraphrase

Repository files navigation

Politeness Paraphrase

This repository contains code and data for the paper: Facilitating the Communication of Politeness through Fine-Grained Paraphrasing. Liye Fu, Susan Fussell and Cristian Danescu-Niculescu-Mizil. EMNLP 2020.

Data

Training and test corpuses are prepared from WikiConv and the Stanford Politeness Corpus and can be found in data (refer to the data README.md for more details).

Training and Evaluation

We include a few notebooks to explain our training and evaluation procedures:

Our pretrained model for adding strategies into messages can be directly downloaded from here (you will need to update the model path in settings.py for some of the notebooks). For reference, generation outputs are provided under:

/outputs
    mt.tsv
    ind.tsv

For a demo on how this approach can be applied to custom texts, see How_to_Make_Strategy_Edits.ipynb.

Dependencies

* convokit=2.4.3
* spacy=2.2.1
* PulP=1.6.8
* GLPK=4.65
* scikit-learn=0.21.3
* dependencies from pytorch_pretrained_bert

Cite as

@InProceedings{fu-paraphrase:2020,
  author={Liye Fu, Susan R. Fussell and Cristian Danescu-Niculescu-Mizil},
  title={Facilitating the Communication of Politeness through Fine-Grained Paraphrasing},
  booktitle={Proceedings of EMNLP},
  year={2020}
}

About

Code and data for the paper "Facilitating the Communication of Politeness through Fine-Grained Paraphrasing". EMNLP 2020.

Resources

Stars

Watchers

Forks