In this thesis we approach the problem of automatically generating naturally sounding sentences. We discover the similarities between a search based realization process and searches in the field of artificial intelligence. These allow us to compile the problem of sentence realization into a representation on which we can use well established techniques from the automatic planning community. We introduce a polynomially space and time bound algorithm and proof its correctness. After various experiments we gathered empirical data, which allows an analysis regarding its practical relevance. Conclusively, we propose ways to further improve the process in the future.