Modern annotation tools should meet at least the following general requirements: they can handle diverse data and annotation levels within one tool, and they support the annotation process with
automatic (pre-)processing outcomes as much as possible. We developed a framework that meets these general requirements and that enables versatile and browser-based annotations of texts, namely
TEXTANNOTATOR. It combines NLP methods of pre-processing with methods of flexible post-processing. In fact, machine learning (ML) requires a lot of training and test data, but is usually far from achieving perfect results. Producing high-level annotations for ML and post-correcting its results are therefore necessary. This is the purpose of TEXTANNOTATOR, which is entirely implemented in ExtJS and provides a range of interactive visualizations of annotations. In addition, it allows for an flexible integration of knowledge resources. The paper describes TEXTANNOTATOR's architecture together with three use cases: annotating temporal structures, argument structures and named entity linking.