Towards Reasonably-Sized Character-Level Transformer NMT by Finetuning Subword Systems

Jindřich Libovický, Alexander Fraser


Abstract
Applying the Transformer architecture on the character level usually requires very deep architectures that are difficult and slow to train. These problems can be partially overcome by incorporating a segmentation into tokens in the model. We show that by initially training a subword model and then finetuning it on characters, we can obtain a neural machine translation model that works at the character level without requiring token segmentation. We use only the vanilla 6-layer Transformer Base architecture. Our character-level models better capture morphological phenomena and show more robustness to noise at the expense of somewhat worse overall translation quality. Our study is a significant step towards high-performance and easy to train character-based models that are not extremely large.
Anthology ID:
2020.emnlp-main.203
Volume:
Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)
Month:
November
Year:
2020
Address:
Online
Venue:
EMNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
2572–2579
Language:
URL:
https://aclanthology.org/2020.emnlp-main.203
DOI:
10.18653/v1/2020.emnlp-main.203
Bibkey:
Cite (ACL):
Jindřich Libovický and Alexander Fraser. 2020. Towards Reasonably-Sized Character-Level Transformer NMT by Finetuning Subword Systems. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2572–2579, Online. Association for Computational Linguistics.
Cite (Informal):
Towards Reasonably-Sized Character-Level Transformer NMT by Finetuning Subword Systems (Libovický & Fraser, EMNLP 2020)
Copy Citation:
PDF:
https://preview.aclanthology.org/paclic-22-ingestion/2020.emnlp-main.203.pdf
Optional supplementary material:
 2020.emnlp-main.203.OptionalSupplementaryMaterial.tgz
Video:
 https://slideslive.com/38938871
Code
 jlibovicky/char-nmt +  additional community code