基于两种新颖辅助任务的端到端语音翻译(End-to-End Speech Translation Enhanced by Two Novel Auxiliary Tasks)

Dou Huaixia (窦怀厦), Lvu Mengzhe (吕孟哲), Li Junhui (李军辉)


Abstract
“端到端语音翻译具有跨模态和跨语言的特性,如何有效地利用这些特性是一个具有挑战性的问题。本文基于多任务学习框架,提出两种新颖辅助任务。语音增强的文本翻译任务通过在文本翻译任务中融入语音模态信息来缓解语音和文本的模态差异,最终提升语音翻译任务的性能。全局感知条件掩码语言建模任务能够同时建模转录文本和译文进而利用文本的全局上下文信息指导翻译模型的训练。在MuST-C数据集8个语向的实验结果表明,本文的方法显著优于基线系统,并且达到了与其它端到端语音翻译系统可竞争的性能水平。进一步的分析实验表明,本文的方法能够缓解语音和文本之间的模态差异并且在不损害文本翻译任务性能的情况下提升语音翻译任务的性能。”
Anthology ID:
2024.ccl-1.13
Volume:
Proceedings of the 23rd Chinese National Conference on Computational Linguistics (Volume 1: Main Conference)
Month:
July
Year:
2024
Address:
Taiyuan, China
Editors:
Maosong Sun, Jiye Liang, Xianpei Han, Zhiyuan Liu, Yulan He
Venue:
CCL
SIG:
Publisher:
Chinese Information Processing Society of China
Note:
Pages:
183–196
Language:
Chinese
URL:
https://aclanthology.org/2024.ccl-1.13/
DOI:
Bibkey:
Cite (ACL):
Dou Huaixia, Lvu Mengzhe, and Li Junhui. 2024. 基于两种新颖辅助任务的端到端语音翻译(End-to-End Speech Translation Enhanced by Two Novel Auxiliary Tasks). In Proceedings of the 23rd Chinese National Conference on Computational Linguistics (Volume 1: Main Conference), pages 183–196, Taiyuan, China. Chinese Information Processing Society of China.
Cite (Informal):
基于两种新颖辅助任务的端到端语音翻译(End-to-End Speech Translation Enhanced by Two Novel Auxiliary Tasks) (Huaixia et al., CCL 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.ccl-1.13.pdf