This paper presents results from a series of simulations that attempted to teach a vanilla sequence-to-sequence neural network a reduplication process. These attempts did not succeed, suggesting that added machinery is necessary for connectionist models to perform such a task.