File Download
There are no files associated with this item.
Supplementary
-
Citations:
- Appears in Collections:
Conference Paper: Improved zero-shot neural machine translation via ignoring spurious correlations
Title | Improved zero-shot neural machine translation via ignoring spurious correlations |
---|---|
Authors | |
Issue Date | 2019 |
Publisher | Association for Computational Linguistics. |
Citation | Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL), Florence, Italy, 28 July - 2 August 2019, p. 1258–1268 How to Cite? |
Abstract | Zero-shot translation, translating between language pairs on which a Neural Machine Translation (NMT) system has never been trained, is an emergent property when training the system in multilingual settings. However, naive training for zero-shot NMT easily fails, and is sensitive to hyper-parameter setting. The performance typically lags far behind the more conventional pivot-based approach which translates twice using a third language as a pivot. In this work, we address the degeneracy problem due to capturing spurious correlations by quantitatively analyzing the mutual information between language IDs of the source and decoded sentences. Inspired by this analysis, we propose to use two simple but effective approaches: (1) decoder pre-training; (2) back-translation. These methods show significant improvement (4 22 BLEU points) over the vanilla zero-shot translation on three challenging multilingual datasets, and achieve similar or better results than the pivot-based approach. |
Description | Poster Session 2B: Machine Translation - no. 7 |
Persistent Identifier | http://hdl.handle.net/10722/277807 |
ISBN |
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Gu, J | - |
dc.contributor.author | Wang, Y | - |
dc.contributor.author | Cho, K | - |
dc.contributor.author | Li, VOK | - |
dc.date.accessioned | 2019-10-04T08:01:44Z | - |
dc.date.available | 2019-10-04T08:01:44Z | - |
dc.date.issued | 2019 | - |
dc.identifier.citation | Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL), Florence, Italy, 28 July - 2 August 2019, p. 1258–1268 | - |
dc.identifier.isbn | 978-1-950737-48-2 | - |
dc.identifier.uri | http://hdl.handle.net/10722/277807 | - |
dc.description | Poster Session 2B: Machine Translation - no. 7 | - |
dc.description.abstract | Zero-shot translation, translating between language pairs on which a Neural Machine Translation (NMT) system has never been trained, is an emergent property when training the system in multilingual settings. However, naive training for zero-shot NMT easily fails, and is sensitive to hyper-parameter setting. The performance typically lags far behind the more conventional pivot-based approach which translates twice using a third language as a pivot. In this work, we address the degeneracy problem due to capturing spurious correlations by quantitatively analyzing the mutual information between language IDs of the source and decoded sentences. Inspired by this analysis, we propose to use two simple but effective approaches: (1) decoder pre-training; (2) back-translation. These methods show significant improvement (4 22 BLEU points) over the vanilla zero-shot translation on three challenging multilingual datasets, and achieve similar or better results than the pivot-based approach. | - |
dc.language | eng | - |
dc.publisher | Association for Computational Linguistics. | - |
dc.relation.ispartof | Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL) | - |
dc.title | Improved zero-shot neural machine translation via ignoring spurious correlations | - |
dc.type | Conference_Paper | - |
dc.identifier.email | Wang, Y: wangyong@eee.hku.hk | - |
dc.identifier.email | Li, VOK: vli@eee.hku.hk | - |
dc.identifier.authority | Li, VOK=rp00150 | - |
dc.identifier.doi | 10.18653/v1/P19-1121 | - |
dc.identifier.hkuros | 306515 | - |
dc.identifier.spage | 1258 | - |
dc.identifier.epage | 1268 | - |
dc.publisher.place | Stroudsburg, PA | - |