While recent advances in deep learning led to significant improvements in\nmachine translation, neural machine translation is often still not able to\ncontinuously adapt to the environment. For humans, as well as for machine\ntranslation, bilingual dictionaries are a promising knowledge source to\ncontinuously integrate new knowledge. However, their exploitation poses several\nchallenges: The system needs to be able to perform one-shot learning as well as\nmodel the morphology of source and target language.\n In this work, we proposed an evaluation framework to assess the ability of\nneural machine translation to continuously learn new phrases. We integrate\none-shot learning methods for neural machine translation with different word\nrepresentations and show that it is important to address both in order to\nsuccessfully make use of bilingual dictionaries. By addressing both challenges\nwe are able to improve the ability to translate new, rare words and phrases\nfrom 30% to up to 70%. The correct lemma is even generated by more than 90%.\n