While automatic emotion detection in text is still in its infancy, machine learning techniques are currently considered as a state-of-the-art methodology to identify emotional states in text. These systems, however, rely on large quantities of annotated data from which prediction models are derived, so progress is mainly made for majority languages such as English. As the system’s performance is dependent on the kind of data it is trained on (e.g. the language, domain and label set used for annotation), every shift to new data implies a new annotation effort. This process is not only arduous, but also inhibits progress in multilingual NLP. The focus in this project is on emotion detection for Dutch, for which currently no annotated data are publicly available, and on exploring transfer learning techniques to allow for cross-language transfer. In transfer learning, the aim is to transfer information from previously learned systems to novel tasks, reducing the needed amount of training data. By reusing information from English systems for Dutch emotion detection, we aim to tackle the data acquisition bottleneck and investigate whether language-independent emotion features can be learned. Moreover, we will investigate transfer across domains and label sets. This will result in a flexible Dutch emotion detection methodology which can be applied to various use cases. We will explore one such case by automatically detecting real-life emotions that emerge during crisis situations.