majumderb / pabst

Code for "Unsupervised Enrichment of Persona-grounded Dialog with Background Stories", ACL 2021
MIT License
10 stars 1 forks source link
commonsense-knowledge decoding dialog unsupervised

Code for "Unsupervised Enrichment of Persona-grounded Dialog with Background Stories"

Unsupervised Enrichment of Persona-grounded Dialog with Background Stories

Bodhisattwa Prasad Majumder, Taylor Berg-Kirkpatrick, Julian McAuley, Harsh Jhamtani

Published at Association for Computational Linguistics (ACL), 2021.

pabst

Data

Contains DNLI dataset to train the persona-consistency model in data/dnli.

Training and Inference

Persona-consistency

The training script for training the persona-consistency classifier here. Use the dnli dataset stated above. It may achieve near-perfect accuracy.

Generation (Inference)

The generation script (here) uses the persona-consistency model as one of the soft constraints.

Citation

If you find pabst useful for your research, please cite our paper:

@inproceedings{MajumderBMJ21,
  author    = {Bodhisattwa Prasad Majumder and
               Taylor Berg{-}Kirkpatrick and
               Julian J. McAuley and
               Harsh Jhamtani},
  title     = {Unsupervised Enrichment of Persona-grounded Dialog with Background Stories},
  booktitle = {ACL},
  year      = {2021},
  url       = {https://arxiv.org/pdf/2106.08364.pdf},
}