We built an NLP system implementing a "virtual writing conference" for elementary-school children, with German as the target language. Currently, state-of-the-art computer support for writing tasks is restricted to multiple-choice questions or quizzes because automatic parsing of the often ambiguous and fragmentary texts produced by pupils presents insurmountable problems. Here, we follow a different course by deploying natural language generation technology to evaluate and improve the grammatical quality of student output. Based on an abstract representation of the story under construction, all paraphrases of simple and combined clauses are generated fully automatically. From this source, the system produces exercises enabling the pupils to improve their sentences. We apply parsing technology only in the teacher mode, where new stories are entered into the system in a simple manner. The system we describe here is a prototype including a full exercise generation mode and a rudimentary teacher mode. (Contains 7 notes, 1 table, and 4 figures.)