Title: Adaptation for Interactive Neural Machine Translation Abstract: (Online-) Adaptation is a key feature of modern computer-aided translation systems. The users of these systems, which are either post-editing machine translation (MT) outputs or are making use of interactive MT, reasonably expect that the underlying MT system will not repeat errors that have been corrected before. Furthermore, the system is also expected to adapt to the general domain of the to be translated text. With the previous phrase-based statistical MT approaches, both of these tasks were daunting due to the range of independent sub-models that had to be adapted. Fortunately, with neural MT, fine-tuning can be readily applied to quickly adapt a model's parameters to new training examples. Interactive MT systems do however have tight latency constraints, which are quickly violated when a personalized model has to be loaded for each user prior to translation. In this talk, we describe methods for efficient and effective adaptation of interactive neural MT systems in a production setting, and present evaluation metrics that enable us to compare adaptation methods in terms of their capability of learning from just a few examples as well as in terms of general domain adaptation performance.