Domain Adaptation and Multi-Domain Adaptation for Neural Machine Translation: A Survey

The development of deep learning techniques has allowed Neural Machine Translation (NMT) models to become extremely powerful, given sufficient training data and time. However, systems struggle when translating text from a new domain with distinct style or vocabulary. Fine-tuning on in-domain allows good adaptation, but requires relevant bilingual data. Even if this is available, simple fine-tuning can cause overfitting catastrophic forgetting previously learned behaviour. We survey approaches adaptation for NMT, particularly where system may need translate across multiple domains. divide into those revolving around selection generation, model architecture, parameter procedure, inference procedure. finally highlight the benefits multidomain other lines NMT research.