Enhancing Cross-lingual Prompting with Dual Prompt Augmentation
This work addresses data scarcity in few-shot cross-lingual prompting for NLP applications, representing an incremental improvement over prior methods.
The paper tackles the under-explored potential of prompting for multilingual problems by proposing a dual prompt augmentation framework to address data scarcity in few-shot cross-lingual scenarios, achieving 46.54% accuracy on XNLI with only 16 English examples per class compared to 34.99% for finetuning.
Prompting shows promising results in few-shot scenarios. However, its strength for multilingual/cross-lingual problems has not been fully exploited. Zhao and Schütze (2021) made initial explorations in this direction by presenting that cross-lingual prompting outperforms cross-lingual finetuning. In this paper, we conduct an empirical exploration on the effect of each component in cross-lingual prompting and derive language-agnostic Universal Prompting, which helps alleviate the discrepancies between source-language training and target-language inference. Based on this, we propose DPA, a dual prompt augmentation framework, aiming at relieving the data scarcity issue in few-shot cross-lingual prompting. Notably, for XNLI, our method achieves 46.54% with only 16 English training examples per class, significantly better than 34.99% of finetuning. Our code is available at https://github.com/DAMO-NLP-SG/DPA.