BIT-Xiaomi’s Simultaneous Translation System for AutoSimTrans 2022

Mengge Liu, Xiang Li, Bao Chen, Yanzhi Tian, Tianwei Lan, Silin Li, Yuhang Guo*, Jian Luan, Bin Wang

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

This system paper describes the BIT-Xiaomi simultaneous translation system for Autosimtrans 2022 simultaneous translation challenge. We participated in three tracks: the Zh-En text-to-text track, the Zh-En audio-to-text track, and the En-Es test-to-text track. In our system, wait-k is utilized to train prefix-to-prefix translation models. We integrate streaming chunking to detect segmentation boundaries as the source streaming reading in. We further improve our system with data selection, data augmentation, and R-Drop training methods. Results show that our wait-k implementation outperforms the organizer’s baseline by at most 8 BLEU score and our proposed streaming chunking method further improves by about 2 BLEU score in the low latency regime.

Original languageEnglish
Title of host publicationAutoSimTrans 2022 - Automatic Simultaneous Translation Challenges, Recent Advances, and Future Directions, Proceedings of the 3rd Workshop
EditorsJulia Ive, Ruiqing Zhang
PublisherAssociation for Computational Linguistics (ACL)
Pages34-42
Number of pages9
ISBN (Electronic)9781955917964
Publication statusPublished - 2022
Event3rd Workshop on Automatic Simultaneous Translation Challenges, Recent Advances, and Future Directions, AutoSimTrans 2022 - Virtual, Online
Duration: 15 Jul 202216 Jul 2022

Publication series

NameAutoSimTrans 2022 - Automatic Simultaneous Translation Challenges, Recent Advances, and Future Directions, Proceedings of the 3rd Workshop

Conference

Conference3rd Workshop on Automatic Simultaneous Translation Challenges, Recent Advances, and Future Directions, AutoSimTrans 2022
CityVirtual, Online
Period15/07/2216/07/22

Fingerprint

Dive into the research topics of 'BIT-Xiaomi’s Simultaneous Translation System for AutoSimTrans 2022'. Together they form a unique fingerprint.

Cite this