The model lacks an end-to-end dataset, and generated poems may not transfer well to Shakespearean style because the style-transfer training data poorly represents poetic language.