Towards Continual Learning for Multilingual Machine Translation via Vocabulary Substitution

We propose a straightforward vocabulary adaptation scheme to extend the\nlanguage capacity of multilingual machine translation models, paving the way\ntowards efficient continual learning for multilingual machine translation. Our\napproach is suitable for large-scale datasets, applies to distant languages\nwith unseen scripts, incurs only minor degradation on the translation\nperformance for the original language pairs and provides competitive\nperformance even in the case where we only possess monolingual data for the new\nlanguages.\n