arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2008.00401cs.CL
收藏

基于可扩展多语言预训练和微调的多语言翻译

Multilingual Translation with Extensible Multilingual Pretraining and Finetuning

Facebook AI(Facebook 人工智能研究院)

Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, Angela Fan

摘要:

最近的工作证明了多语言预训练的潜力,即创建一个可用于不同语言的各种任务的模型。以前在多语言预训练方面的工作已经证明,机器翻译系统可以通过对bitext进行微调来创建。在这项工作中,我们展示了可以通过多语言微调创建多语言翻译模型。预先训练的模型不是在一个方向上进行微调,而是同时在多个方向上进行微调。与从头开始训练的多语言模型相比,从预训练模型开始融合了大量未标记单语数据的好处,这对于没有双文本的低资源语言尤为重要。我们证明了预训练的模型可以扩展到包含其他语言而不会损失性能。我们将mBART中的语言数量增加了一倍,以支持50种语言的多语言机器翻译模型。最后,我们创建了ML50基准测试,涵盖了低、中和高资源语言,通过标准化训练和评估数据来促进可重复的研究。在ML50上,我们证明了多语言微调比最强的基线(无论是从头开始的多语言还是双语微调)平均提高了1 BLEU,而从从头开始的双语基线平均提高了9.3 BLEU。

英文摘要:

Recent work demonstrates the potential of multilingual pretraining of creating one model that can be used for various tasks in different languages. Previous work in multilingual pretraining has demonstrated that machine translation systems can be created by finetuning on bitext. In this work, we show that multilingual translation models can be created through multilingual finetuning. Instead of finetuning on one direction, a pretrained model is finetuned on many directions at the same time. Compared to multilingual models trained from scratch, starting from pretrained models incorporates the benefits of large quantities of unlabeled monolingual data, which is particularly important for low resource languages where bitext is not available. We demonstrate that pretrained models can be extended to incorporate additional languages without loss of performance. We double the number of languages in mBART to support multilingual machine translation models of 50 languages. Finally, we create the ML50 benchmark, covering low, mid, and high resource languages, to facilitate reproducible research by standardizing training and evaluation data. On ML50, we demonstrate that multilingual finetuning improves on average 1 BLEU over the strongest baselines (being either multilingual from scratch or bilingual finetuning) while improving 9.3 BLEU on average over bilingual baselines from scratch.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑