This article presents briefly the advantages and disadvantages of reusing natural language processing (NLP) tools in the CALL context. The issue is addressed through the description of GBGen, a sentence generation system. The abstract semantic representation used as input for the generator is described, as well as the actual generation process, from a deep structure to a grammatical sentence through transformations and the application of morphology. The possible didactic value of such a tool is then evaluated and the outline of a CALL scenario given. Finally, proposed adaptations of the generator for the CALL context are discussed.