This paper presents Coral, an interface in which complex corpus queries can be expressed in a controlled subset of natural English. With the help of a predictive editor, users can compose queries and submit them to the Coral system, which then automatically translates them into formal AQL statements. The paper gives an overview of the controlled natural language developed for Coral and describes the functionalities of the predictive editor provided for it. It also reports on a user experiment in which the system was evaluated. The results show that, with Coral, corpora of annotated texts can be queried easier and faster than with the existing ANNIS interface. Our system demonstrates that complex corpora can be accessed without the need to learn a complicated formal query language.