This bachelor thesis compares the long short-term memory (LSTM) unit and a recently proposed gated recurrent unit (GRU) in a empirical way. The comparison task is the sequence-to-sequence model, which has its roots in the natural language generation field. The task of the sequence-to-sequence model is to generate from a triple value an corresponding natural language sentence. The experiments in this bachelor thesis revealed that there is no superior unit and both units have advantages in certain scenarios.