Abstract
This paper introduces the overall situation of the Natural Language Processing Laboratory of Tianjin University participating in the WMT23 machine translation evaluation task from Chinese to English. For this evaluation, the base model used is a Transformer based on a Mixture of Experts (MOE) model. During the model’s construction and training, a basic dense model based on Transformer is first trained on the training set. Then, this model is used to initialize the MOE-based translation model, which is further trained on the training corpus. Since the training dataset provided for this translation task is relatively small, to better utilize sparse models to enhance translation, we employed a data augmentation technique for alignment. Experimental results show that this method can effectively improve neural machine translation performance.- Anthology ID:
- 2023.wmt-1.33
- Volume:
- Proceedings of the Eighth Conference on Machine Translation
- Month:
- December
- Year:
- 2023
- Address:
- Singapore
- Editors:
- Philipp Koehn, Barry Haddow, Tom Kocmi, Christof Monz
- Venue:
- WMT
- SIG:
- SIGMT
- Publisher:
- Association for Computational Linguistics
- Note:
- Pages:
- 307–311
- Language:
- URL:
- https://aclanthology.org/2023.wmt-1.33
- DOI:
- 10.18653/v1/2023.wmt-1.33
- Cite (ACL):
- Shaolin Zhu and Deyi Xiong. 2023. TJUNLP:System Description for the WMT23 Literary Task in Chinese to English Translation Direction. In Proceedings of the Eighth Conference on Machine Translation, pages 307–311, Singapore. Association for Computational Linguistics.
- Cite (Informal):
- TJUNLP:System Description for the WMT23 Literary Task in Chinese to English Translation Direction (Zhu & Xiong, WMT 2023)
- PDF:
- https://preview.aclanthology.org/proper-vol2-ingestion/2023.wmt-1.33.pdf