Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Consistency Regularization for Cross-Lingual Fine-Tuning

About

Fine-tuning pre-trained cross-lingual language models can transfer task-specific supervision from one language to the others. In this work, we propose to improve cross-lingual fine-tuning with consistency regularization. Specifically, we use example consistency regularization to penalize the prediction sensitivity to four types of data augmentations, i.e., subword sampling, Gaussian noise, code-switch substitution, and machine translation. In addition, we employ model consistency to regularize the models trained with two augmented versions of the same training set. Experimental results on the XTREME benchmark show that our method significantly improves cross-lingual fine-tuning across various tasks, including text classification, question answering, and sequence labeling.

Bo Zheng, Li Dong, Shaohan Huang, Wenhui Wang, Zewen Chi, Saksham Singhal, Wanxiang Che, Ting Liu, Xia Song, Furu Wei• 2021

Related benchmarks

TaskDatasetResultRank
Named Entity RecognitionCoNLL (test)
F1 Score (AVG)78.85
35
Question AnsweringXTREME QA Order 1 - Short Sequence (test)
EM56.96
16
Question AnsweringXTREME QA Order 2 - Short Sequence (test)
EM56.96
16
Question AnsweringXTREME QA Order 3 - Short Sequence (test)
EM56.96
16
Question AnsweringXTREME QA Order 4 - Long Sequence (test)
EM59.93
16
Question AnsweringXTREME QA Order 5 - Long Sequence (test)
EM59.93
16
Question AnsweringXTREME QA Order 6 - Long Sequence (test)
EM0.5993
16
Question AnsweringXTREME QA Average across all orders (test)
EM58.45
16
Showing 8 of 8 rows

Other info

Follow for update