This repository contains the code and data for our NAACL 2022 paper:
Uncertainty-Aware Cross-Lingual Transfer with Pseudo Partial Labels
If you find this work useful and use it on your own research, please cite our paper.
@inproceedings{lei-etal-2022-uncertainty,
title = {Uncertainty-Aware Cross-Lingual Transfer with Pseudo Partial Labels},
author = {Lei, Shuo and
Zhang, Xuchao and
He, Jianfeng and
Chen, Fanglan and
Lu, Chang-Tien},
booktitle = {Findings of the Association for Computational Linguistics: NAACL 2022},
month = jul,
year = {2022},
publisher = {Association for Computational Linguistics},
pages = {1987--1997},
}
In this work, we propose an uncertainty-aware Cross-Lingual Transfer framework with Pseudo-Partial-Label (CLTP) to maximize the utilization of unlabeled data by reducing the noise introduced in the training phase. To estimate pseudo-partial-label for each unlabeled data, we propose a novel estimation method, considering both prediction confidence and the limitation to the number of similar labels.
The figure gives an illustration of our method. For an unlabeled sample with ambiguous predictions, the standard one-hot-labeling takes the class with the highest confidence as the pseudo-one-hot-label, introducing the noise in the training phase due to the wrong prediction. Instead of choosing one among the predictions that all have low confidence, the proposed partial-labeling method takes both ambiguous classes as candidate labels, allowing the ground-truth label to be presented in the training phase.
We ran experiments on a total of 15 languages. You may download data file by running the following commands and put the data files under download/ folder.
cd bin
sh download_data.sh
After you have finished configuring the download/ folder, you can run our model with the following commands.
cd bin
sh xnli.sh
You can also adjust the model by modifying the parameters in the experiments.conf file.
- Python 3.7
- PyTorch 1.9.1
- transformer 4.17.3
- numpy 1.19.1
- tqdm 4.62.3
- CUDA 11.4