Abstract
Neural Machine Translation (NMT) approaches employing monolingual data are showing steady improvements in resource-rich conditions. However, evaluations using real-world lowresource languages still result in unsatisfactory performance. This work proposes a novel zeroshot NMT modeling approach that learns without the now-standard assumption of a pivot language sharing parallel data with the zero-shot source and target languages. Our approach is based on three stages: initialization from any pre-trained NMT model observing at least the target language, augmentation of source sides leveraging target monolingual data, and learning to optimize the initial model to the zero-shot pair, where the latter two constitute a selflearning cycle. Empirical findings involving four diverse (in terms of a language family, script and relatedness) zero-shot pairs show the effectiveness of our approach with up to +5.93 BLEU improvement against a supervised bilingual baseline. Compared to unsupervised NMT, consistent improvements are observed even in a domain-mismatch setting, attesting to the usability of our method.- Anthology ID:
- 2021.mtsummit-loresmt.10
- Volume:
- Proceedings of the 4th Workshop on Technologies for MT of Low Resource Languages (LoResMT2021)
- Month:
- August
- Year:
- 2021
- Address:
- Virtual
- Venue:
- LoResMT
- SIG:
- Publisher:
- Association for Machine Translation in the Americas
- Note:
- Pages:
- 96–113
- Language:
- URL:
- https://aclanthology.org/2021.mtsummit-loresmt.10
- DOI:
- Cite (ACL):
- Surafel M. Lakew, Matteo Negri, and Marco Turchi. 2021. Zero-Shot Neural Machine Translation with Self-Learning Cycle. In Proceedings of the 4th Workshop on Technologies for MT of Low Resource Languages (LoResMT2021), pages 96–113, Virtual. Association for Machine Translation in the Americas.
- Cite (Informal):
- Zero-Shot Neural Machine Translation with Self-Learning Cycle (Lakew et al., LoResMT 2021)
- PDF:
- https://preview.aclanthology.org/auto-file-uploads/2021.mtsummit-loresmt.10.pdf