Visual Cues and Error Correction for Translation Robustness

Zhenhao Li, Marek Rei, Lucia Specia


Abstract
Neural Machine Translation models are sensitive to noise in the input texts, such as misspelled words and ungrammatical constructions. Existing robustness techniques generally fail when faced with unseen types of noise and their performance degrades on clean texts. In this paper, we focus on three types of realistic noise that are commonly generated by humans and introduce the idea of visual context to improve translation robustness for noisy texts. In addition, we describe a novel error correction training regime that can be used as an auxiliary task to further improve translation robustness. Experiments on English-French and English-German translation show that both multimodal and error correction components improve model robustness to noisy texts, while still retaining translation quality on clean texts.
Anthology ID:
2021.findings-emnlp.271
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2021
Month:
November
Year:
2021
Address:
Punta Cana, Dominican Republic
Venues:
EMNLP | Findings
SIG:
SIGDAT
Publisher:
Association for Computational Linguistics
Note:
Pages:
3153–3168
Language:
URL:
https://aclanthology.org/2021.findings-emnlp.271
DOI:
10.18653/v1/2021.findings-emnlp.271
Bibkey:
Cite (ACL):
Zhenhao Li, Marek Rei, and Lucia Specia. 2021. Visual Cues and Error Correction for Translation Robustness. In Findings of the Association for Computational Linguistics: EMNLP 2021, pages 3153–3168, Punta Cana, Dominican Republic. Association for Computational Linguistics.
Cite (Informal):
Visual Cues and Error Correction for Translation Robustness (Li et al., Findings 2021)
Copy Citation:
PDF:
https://aclanthology.org/2021.findings-emnlp.271.pdf
Code
 nickeilf/visual-cues-error-correction