Abstract:While NMT has achieved remarkable results in the last 5 years, production systems come with strict quality requirements in arbitrarily niche domains that are not always adequately covered by readily available parallel corpora. This is typically addressed by training domain specific models, using fine-tuning methods and some variation of back-translation on top of in-domain monolingual corpora. However, industrial practitioners can rarely afford to focus on a single domain. A far more typical scenario includes a set of closely related, yet succinctly different sub-domains. At <a class="link-external link-http" href="http://Booking.com" rel="external noopener nofollow">this http URL</a>, we need to translate property descriptions, user reviews, as well as messages, (for example those sent between a customer and an agent or property manager). An editor might need to translate articles across a set of different topics. An e-commerce platform would typically need to translate both the description of each item and the user generated content related to them. To this end, we propose MDT: a novel method to simultaneously fine-tune on several sub-domains by passing multidimensional sentence-level information to the model during training and inference. We show that MDT achieves results competitive to N specialist models each fine-tuned on a single constituent domain, while effectively serving all N sub-domains, therefore cutting development and maintenance costs by the same factor. Besides BLEU (industry standard automatic evaluation metric known to only weakly correlate with human judgement) we also report rigorous human evaluation results for all models and sub-domains as well as specific examples that better contextualise the performance of each model in terms of adequacy and fluency. To facilitate further research, we plan to make the code available upon acceptance.

Domain Sensitive Chinese-English Example Based Machine Translation

Auto Word Alignment Based Chinese-English EBMT.

Example-Based Chinese-English Mt

Example-based Machine Translation Based on Tree–string Correspondence and Statistical Generation

Feature Rich Translation Model for Example-Based Machine Translation

A Hybrid Approach to Chinese-English Machine Translation

Implementation of Chinese-Uyghur Bilateral EBMT System

Maximum Entropy Model for Example-Based Machine Translation

Multilingual Machine Translation system

Combining Domain Adaptation Approaches for Medical Text Translation

Extraction of Translation Example Based on Shallow Parsing Information

Domain Adaptation for Statistical Machine Translation

English-Chinese Bilingual Teaching: A DECTMT-NBO-DMSFNN Approach for Design and Application of Machine Translation Technology

CHINESE-ENGLISH MACHINE TRANSLATION DISAMBIGUATING WITH RULE-BASED METHOD COMBINED WITH STATISTIC-BASED METHOD

Domain adaptation for statistical machine translation with domain dictionary and monolingual corpora

Chinese-English Smt for Cross-Language Dialogue Agent Support

Multi-Domain Neural Machine Translation with Word-Level Domain Context Discrimination

Domain Adaptation for Medical Text Translation Using Web Resources

Bridging the Domain Gap: Improve Informal Language Translation Via Counterfactual Domain Adaptation

Multi-Domain Adaptation in Neural Machine Translation Through Multidimensional Tagging

Domain-specific machine translation with recurrent neural network for software localization