Enhancing Cross-lingual Prompting with Dual Prompt Augmentation

This is a PyTorch implementation of our paper at ACL 2023 Findings: "Enhancing Cross-lingual Prompting with Dual Prompt Augmentation".

Shell Scripts

• dpa-{xnli,pawsx}.sh: shell scripts to run DPA training, different strategies can be played around by modifying parameters
• {xnli,paws-x}.sh shell scripts to run finetuning as a baseline

Python Codes

• run_{xnli,pawsx}.py: python code for finetuning, modified from https://github.com/huggingface/transformers/blob/main/examples/pytorch/text-classification/run_xnli.py
• run-prompt.py: python code for prompting with DPA
• balanced_data_processor.py: for sampling a balanced training/dev set for few-shot experiments
• prompt_helper.py: some helper utils for prompting
• modeling_xlmr.py: the model class for prompting XLM-R
• xnli-metrics.py: an off-the-shelf script from datasets

Dependencies

• transformers==4.10.3
• datasets==1.12.1
• torch==1.7.1

Citation

If you find our code useful, please cite the following, thank you!

```bibtxt
@inproceedings{acl23/DPA,
  author    = {Meng Zhou and
               Xin Li and
               Yue Jiang and
               Lidong Bing},
  title     = {Enhancing Cross-lingual Prompting with Dual Prompt Augmentation},
  booktitle = {Findings of the 2023 ACL},
  year      = {2023},
  url       = {},
}
```